PaperlayerView exact revision on arXiv

Source-linked AI summary

A Survey on Resource Management in Joint Communication and Computing-Embedded SAGIN

Qian Chen, Zheng Guo, Weixiao Meng, Shuai Han, Cheng Li, Tony Q. S. Quek

arXiv:2403.17400v3cs.NIcs.DC

TL;DR

Remote areas remain beyond terrestrial coverage, while latency-sensitive and computation-intensive services motivate JCC-SAGIN. This paper reviews its architecture, technologies, applications, and resource-management methods, then outlines future research directions.

Abstract

from arXiv · show

The advent of the 6G era aims for ubiquitous connectivity, with the integration of non-terrestrial networks (NTN) offering extensive coverage and enhanced capacity. As manufacturing advances and user demands evolve, space-air-ground integrated networks (SAGIN) with computational capabilities emerge as a viable solution for services requiring low latency and high computational power. Resource management within joint communication and computing-embedded SAGIN (JCC-SAGIN) presents greater complexity than traditional terrestrial networks. This complexity arises from the spatiotemporal dynamics of network topology and service demand, the interdependency of large-scale resource variables, and intricate tradeoffs among various performance metrics. Thus, a thorough examination of resource management strategies in JCC-SAGIN is crucial, emphasizing the role of non-terrestrial platforms with processing capabilities in 6G. This paper begins by reviewing the architecture, enabling technologies, and applications in JCC-SAGIN. Then, we offer a detailed overview of resource management modeling and optimization methods, encompassing both traditional optimization approaches and learning-based intelligent decision-making frameworks. Finally, we outline the prospective research directions in JCC-SAGIN.

I. INTRODUCTION

JCC-SAGIN combines onboard-capable non-terrestrial platforms with communication and computing to address coverage and service needs. Existing SAGIN surveys leave the effects of SAP computing on performance underexplored.

A. Background and Motivation

  • Terrestrial networks cover only 7% of Earth’s surface, leaving an estimated 3.4 billion people without access; 6G envisions satellites, UAVs, and HAPs in SAGIN to extend coverage.The paper refers to non-terrestrial platforms as space-air platforms (SAPs).Source: A. Background and Motivation · p. 1
  • Onboard processing makes communication-computing integration in NTN feasible, allowing visible SAPs to edge-process latency-sensitive or computation-intensive tasks in remote areas.These services would otherwise be constrained by terrestrial IoT devices’ computing capacity or rely on remote ground-station cloud computing.Source: A. Background and Motivation · pp. 1–2

1) Surveys on NTN:

  • Recent NTN surveys have shifted from single-layer networks toward multilayer frameworks and from communication-only topics toward communication and computing, but SAGIN surveys still leave SAP computing’s performance impact underexplored.The identified gap motivates examining computing functions alongside communication in SAGIN.Source: 1) Surveys on NTN · p. 2

2) Surveys on the Integration of Communication with Other Technologies:

  • JCC resource management extends beyond communication resources and QoS to computing metrics, processing and storage allocation, task offloading, and edge-computing architecture.Offloading decisions depend on node computing capability, communication delay, and resource availability; NTN nodes also face power and computation constraints.Source: 1) Surveys on NTN · p. 2, 2) Surveys on the Integration of… · p. 5
  • Surveys integrating communication with other technologies cover ISAC, IoT sensing-computing, MEC, and networking, caching, and computing integration.These works address signal design, processing and optimization, MEC decisions and resource allocation, and integration tradeoffs.Source: 2) Surveys on the Integration of… · pp. 2–3
  • SAGIN JCC also contends with SAP mobility and long propagation delays, vulnerable LoS channels and impairments, and scalability and interoperability demands.These conditions affect protocols, routing, handover, resource allocation, link management, and network integration.Source: 2) Surveys on the Integration of… · pp. 3, 5

3) Surveys on Resource Management:

The survey identifies a gap in comprehensive resource-management coverage for JCC-SAGIN, then reviews its architecture, enabling technologies, applications, optimization methods, and future research directions.

3) Surveys on Resource Management:

  • Existing surveys address resource management across several domains, but none comprehensively reviews SAGIN across hardware, application scenarios, traditional optimization, and AI decision-making.Source: 3) Surveys on Resource Management · p. 5

4) Contributions of This Paper:

  • JCC-SAGIN centers on non-terrestrial platforms with onboard processing and data forwarding, and the survey reviews the evolution of computing hardware and network integration.Source: 4) Contributions of This Paper · p. 5
  • The survey analyzes enabling technologies and discusses supporting applications, including engineering projects that illustrate their practical implementation.Source: 4) Contributions of This Paper · p. 5
  • Its resource-management review covers modeling and optimization, including traditional techniques, learning-based decision-making, and future directions from the physical layer to the application layer.Source: 4) Contributions of This Paper · p. 5

C. Organization of This Survey

  • The survey proceeds from architecture and enabling technologies to applications, then resource-management modeling and optimization, future opportunities, and a concluding summary.Source: C. Organization of This Survey · pp. 5, 7

II. ARCHITECTURE AND EVOLUTION OF JCC-SAGIN

  • The architecture discussion covers network components, computing-hardware development, and the evolution of network integration in JCC-SAGIN.Source: §II · p. 7

A. Network Components in JCC-SAGIN

JCC-SAGIN combines satellite, aerial, and terrestrial components with distinct coverage, mobility, and computing characteristics. Their capabilities and constraints shape access, communication, and task offloading.

1) Space Segment:

  • GEO satellites offer broad coverage but face propagation delay, attenuation, and scarce synchronous orbit resources; MEO and LEO trade coverage and lower delay against their own orbital constraints.The survey describes hierarchical GEO, MEO, and LEO satellite networks; LEO satellites have lower propagation delay and attenuation but rapid relative motion creates Doppler shifts.Source: 1) Space Segment · p. 7
  • Satellite functionality is evolving from transparent forwarding to onboard processing, while LEO constellations also support direct communication with mobile phones.The survey cites Starlink's onboard computers and describes direct-to-phone initiatives including AST SpaceMobile, Lynk Global, and Starlink V2.Source: 1) Space Segment · p. 7

2) Air Segment:

  • Air platforms include UAVs, civil airplanes, and HAPs; UAVs can act as intelligent computing platforms and extend coverage, but limited energy constrains large-scale service.Their low altitude reduces propagation delays; UAVs also support data collection, processing, distribution, and emergency response.Source: 2) Air Segment · pp. 7–9
  • Civil airplanes offer substantial coverage and fewer load and energy constraints than UAVs, but their routes are less flexible and less stable than satellite trajectories.Source: 2) Air Segment · p. 9

3) Ground Segment:

  • Ground infrastructure includes terrestrial networks with robust base-station computing and support facilities that connect satellite and aircraft platforms to the core network.Satellite ground stations and civil-aircraft gateways exchange measurement, control, and data signals, using high-speed optical connections to the core network.Source: 3) Ground Segment · p. 9
  • Compact handheld and IoT terminals have limited transmission capacity and energy, while visible SAPs can tailor access, transmission, offloading, and backhaul to task QoS.Source: 3) Ground Segment · p. 9

B. Evolution of Computing Hardware in JCC-SAGIN

Advances in aerospace and AI processors are expanding the computing capacity available to SAGIN, while network evolution brings together communication, computing, and increasingly integrated services. The section surveys processor types, offloading approaches, and the progression toward integrated network operation.

B. Evolution of Computing Hardware in JCC-SAGIN

  • Chip advances and broader use of commercial off-the-shelf products are strengthening onboard computing, supporting a shift from satellite repeater operation toward combined communication and computing.The section also notes growing global computing capacity and demand for edge processing as IoT devices and data expand.Source: B. Evolution of Computing Hardware in… · p. 9

1) High-Performance Spaceflight Computing (HPSC) Processors:

2) AI Processors for Space, Air and Ground Segments:

  • AI processors trade off efficiency, energy use, and flexibility: GPUs suit parallel computing, FPGAs offer reconfigurable low-power processing, and ASICs can outperform CPUs and GPUs in specific fields.The section identifies CPUs, GPUs, FPGAs, and ASICs as the four primary AI-computing chip types and describes recent GPU advances.Source: 2) AI Processors for Space, Air and… · p. 10

C. Evolution of Network Integration in JCC-SAGIN

  • Resource-constrained devices can offload computational tasks to satellites, UAVs, or ground servers through MEC, with RIS and energy-harvesting technologies also discussed as network enablers.The section also points to digital twins and centralized or decentralized learning techniques for monitoring, control, resource allocation, and optimization.Source: C. Evolution of Network Integration in… · pp. 11–12
  • Network integration progresses from coverage and service integration through user and organization integration to system integration, with standards, white papers, and industry trials addressing integration across segments.System integration is described as supporting flexible networking, dynamic multilayer routing, coordinated resource scheduling, and seamless transitions between NTN and terrestrial networks.Source: C. Evolution of Network Integration in… · p. 12, Figure 7 · p. 12, Figure 8 · p. 13

D. Summary Remarks

JCC-SAGIN combines heterogeneous platforms and computing hardware with technologies intended to improve network efficiency. The section reviews slicing and software-based control, energy supply and transfer approaches, and challenges that remain for deployment.

D. Summary Remarks

  • JCC-SAGIN brings together satellites, UAVs, and aircraft with aerospace-grade CPUs and AI accelerators, but deployment must contend with Doppler shifts, changing topologies, and nonground-channel characteristics.The section also describes RIS-assisted offloading, energy harvesting, digital twins, and learning-based architectures as technologies being integrated into the network.Source: D. Summary Remarks · p. 14

A. Network Slicing, SDN, and NFV Supported JCC-SAGIN

  • Network slicing tailors virtual networks to service requirements, while SDN and NFV support flexible resource allocation; placement and management of network functions and cross-region data privacy remain challenges.The section describes full, partial, and separated slicing strategies, including separated slicing as a way to mitigate SAP dynamics' effects on service continuity.Source: A. Network Slicing, SDN, and NFV… · pp. 14–15

B. Energy Harvesting and Wireless Power Transfer Enhanced JCC-SAGIN

  • Energy harvesting and wireless power transfer are presented as ways to extend device battery life and support network operation where charging is impractical.UAV-assisted wireless-powered MEC studies optimize objectives including task delay, UAV energy use, packet loss, and user computation rate.Source: B. Energy Harvesting and Wireless… · p. 15
  • A hybrid active-passive communication scheme outperforms using only active communication or backscatter in the weighted sum of computation bits in a UAV-enabled wireless-powered MEC network.Backscatter is described as more energy-efficient than active communication but with a lower offloading rate.Source: B. Energy Harvesting and Wireless… · p. 15
  • The section identifies HAPs, especially larger ones with large solar panels, as more promising than drones for powering IoT terminals.It describes solar energy as supporting HAP flight, communication, computing, and other functions.Source: B. Energy Harvesting and Wireless… · pp. 15–16

C. Reconfigurable Intelligent Surface-Assisted JCC-SAGIN

RIS can modify signals to support coverage and communication, and RIS-assisted MEC workflows use platform selection, resource allocation, and surface configuration to facilitate task offloading. The section outlines satellite and aerial use cases alongside deployment constraints.

C. Reconfigurable Intelligent Surface-Assisted JCC-SAGIN

  • RIS can alter incoming signal amplitude and phase, and may either relay signals in an existing system or initiate communication by transmitting phase-shifted signals using an adjacent RF source.The section notes that ground RIS deployments can face installation-location, cost, and multiple-reflection challenges.Source: C. Reconfigurable Intelligent… · p. 16
  • In cooperative MEC, users select an offloading platform when local resources are insufficient, MEC platforms optimize scheduling and resource allocation, and RIS configuration supports the offloading link.The selected satellite, aerial vehicle, or terrestrial station processes the task and returns results to the user over the RIS-assisted link.Source: C. Reconfigurable Intelligent… · p. 16

1) RIS-Assisted Satellite Communications:

  • Satellite-mounted RIS units can act as passive reflectors, but direct satellite deployments remain uncommon and controller placement trades terrestrial interference avoidance against satellite processing burden.Space RIS deployment also faces configuration overhead and untested durability and effectiveness, with limited opportunities to update or replace units in orbit.Source: 1) RIS-Assisted Satellite Communications · p. 16

2) RIS-Assisted Aerial Communications:

  • Aerial platforms offer flexible RIS installation and higher-altitude line-of-sight opportunities, while UAV mobility increases channel-estimation and reconfiguration overhead and payload limits restrict RIS size.The section presents aerial RIS as a way to provide adaptable wireless coverage.Source: 2) RIS-Assisted Aerial Communications · pp. 16–17

3) RIS-Assisted Cooperative MEC framework in JCC-SAGIN:

The section surveys RIS-assisted MEC and digital-twin approaches in JCC-SAGIN, then describes enabling technologies and application examples. It highlights both resource-management opportunities and constraints on deployment.

3) RIS-Assisted Cooperative MEC framework in JCC-SAGIN:

  • RIS-assisted MEC combines communication-link improvement with computational support, while aerial implementations face resource-management, trajectory, beamforming, and path-loss challenges.Reported simulations found an 8.08 Mb computation-capacity gain over operation without RIS for an OMA scheme; NOMA achieved higher computation capacity than OMA.Source: 3) RIS-Assisted Cooperative MEC… · p. 17

D. Digital Twin-Enabled JCC-SAGIN

  • Digital twins represent physical entities, monitor network status, and can support prediction and intelligent control through two-way interaction with the physical network.Reviewed DT-enabled work includes offline DRL training, DT-integrated federated learning, and decentralized incentives; SAP compute limits motivate lightweight AI models.Source: D. Digital Twin-Enabled JCC-SAGIN · pp. 17–18

E. Summary Remarks

  • Network slicing, energy harvesting and wireless power transfer, RIS, and digital twins support flexible services, sustainable operation, offloading, and resource management.The section presents these technologies as complementary means of addressing SAGIN challenges.Source: E. Summary Remarks · p. 18

A. Water Monitoring

  • Water monitoring combines sensing devices, edge devices, and a central platform for unified data management and collaborative analysis with rapid edge response.The section also describes GRACE satellite observations of gravity-field changes used to assess watershed water reserves and track water movement.Source: A. Water Monitoring · p. 18

B. Smart Oilfields

C. Precision Agriculture

The section surveys NTN applications in precision agriculture, wildlife tracking, and disaster management. Examples range from satellite and UAV sensing to communications and post-disaster observation.

C. Precision Agriculture

  • Precision agriculture integrates IoT, sensors, satellites, and aerial platforms to collect geospatial data for crop monitoring and farming decisions.UAVs support weed control, yield estimation, disease detection, irrigation, and crop spraying; satellite data informs seasonal decisions.Source: C. Precision Agriculture · p. 19

1) Migration Monitoring of Avian Species:

  • Satellite transmitters track bird migrations, helping researchers identify routes and study challenges including habitat degradation and climate-change impacts.The resulting data can inform conservation strategies.Source: 1) Migration Monitoring of Avian Species · p. 20

2) Monitoring of Marine Fauna:

  • Satellite tags provide real-time positional data on marine wildlife, revealing migratory paths, foraging areas, and breeding sites.These observations inform measures such as marine protected areas and fishing regulation.Source: 2) Monitoring of Marine Fauna · p. 20

3) Surveillance of Endangered Species:

  • GPS or satellite collars track endangered animals’ ranges, movements, and habitat preferences to inform conservation and antipoaching strategies.Examples include elephants, tigers, and rhinos.Source: 3) Surveillance of Endangered Species · p. 20

1) Early Warning Through Remote Sensing:

  • Remote sensing supplies hazard information for early warnings, supporting timely alerts and evacuation strategies.The section identifies storms, floods, and wildfires as example hazards.Source: 1) Early Warning Through Remote Sensing · p. 20

E. Disaster Management

  • When terrestrial networks are compromised, satellite phones and data links support disaster rescue, relief coordination, and medical-supply distribution.Drones also survey damage, locate survivors, and deliver supplies, while satellite navigation aids search-and-rescue teams.Source: 2) Emergency Response · p. 20

3) Earth Observation for Post-Disaster Recovery:

  • Satellite imagery supports post-disaster recovery monitoring by informing assessments of reconstruction, relief effectiveness, and areas needing further assistance.The section also notes DSCOVR solar-wind monitoring and GPM precipitation observations for space-weather warnings and disaster alerts.Source: 3) Earth Observation for Post-Disaster… · p. 20

F. Search and Rescue

The section moves from search-and-rescue examples to resource-management modeling in JCC-SAGIN. It reviews common performance metrics and the trade-offs and indicators used to evaluate systems.

1) Damage Assessment and Relief Operations:

  • Disaster-zone drones provide real-time damage and survivor information to guide response priorities and resource allocation.They can also deliver essential supplies to hard-to-reach areas.Source: 1) Damage Assessment and Relief… · p. 20

2) Navigation for Search and Rescue Operations:

  • Satellite navigation supports search-and-rescue coordination, while Cospas-Sarsat uses LEO and GEO satellites, ground stations, and control centers to relay distress signals and locations.The passage also describes Starlink as a rapid disaster-relief communication network when ground infrastructure is compromised.Source: 2) Navigation for Search and Rescue… · p. 20

G. Summary Remarks

  • JCC-SAGIN applications require platforms to provide computation as well as communication, making resource management important for performance, cost, and user-demand alignment.The listed resource choices include user association, power and computing allocation, beamforming, and UAV trajectory optimization.Source: G. Summary Remarks · pp. 20–21

V. RESOURCE MANAGEMENT MODELING IN JCC-SAGIN

  • The modeling overview covers metrics, offloading strategies, network models, and service models as foundations for later optimization methods..Source: §V · p. 21

A. Metric for JCC-SAGIN

  • Task completion time and energy consumption are central metrics, and weighted delay-energy objectives allow scenario-specific trade-offs.Task completion time includes transmission and computing time; satellite-ground propagation makes return-transmission time relevant in JCC-SAGIN.Source: A. Metric for JCC-SAGIN · p. 21, 1) Task Completion Time · p. 21

2) Energy Consumption:

  • Energy-focused studies account for MEC-server consumption and constrained satellite and aerial resources, including through offloading and resource-allocation designs.The reviewed work includes delay-constrained energy minimization and a three-layer LEO satellite offloading architecture.Source: 2) Energy Consumption · p. 21

3) Weighted Sum of Delay and Energy Consumption:

  • Weighted sums of task completion time and energy consumption let systems tune the relative importance of delay and energy under resource constraints.Reviewed examples include LEO satellite computing and multi-tier local, SAP-edge, and ground-cloud computing.Source: 3) Weighted Sum of Delay and Energy… · p. 22

4) Other Key Performance Indicators:

  • Other indicators include user experience, input bits processed, information freshness, and computational reliability.Metric selection varies with the research problem and system architecture.Source: 4) Other Key Performance Indicators · p. 22

B. Offloading Strategy in JCC-SAGIN

The section reviews offloading choices, NTN performance-analysis methods, and service modeling. It emphasizes that JCC-SAGIN models must account for platform distributions, task attributes, and spatiotemporal service variation.

B. Offloading Strategy in JCC-SAGIN

  • Partial offloading splits independent task bits across platforms for parallel processing, while binary offloading assigns the whole task to one platform.Reviewed comparisons report typically longer completion time for binary offloading and advantages for partial offloading in sum computation bits.Source: B. Offloading Strategy in JCC-SAGIN · p. 22

1) Platform Distribution:

  • A finite number of dispersed SAPs is better represented by a binomial point process than by the terrestrial-network Poisson point process in the described settings.Reviewed NTN studies model UAV and LEO satellite locations with three-dimensional BPPs and also use nonhomogeneous PPPs for latitude-dependent satellite analysis.Source: 1) Platform Distribution · pp. 22–23

2) Performance Analysis:

  • NTN performance analysis focuses mainly on coverage and capacity, using simulation or numerical and theoretical methods.Simulation studies report reduced delay and packet loss with more satellites or inter-satellite links, and diminishing capacity gains as constellation scale grows; system-level analysis under SR fading remains limited.Source: 2) Performance Analysis · p. 23

1) Service Attributes:

  • Task attributes commonly include input size, required CPU cycles, and tolerable completion delay, with output-to-input size ratios sometimes used to describe returned results.Some studies omit return transmission when the output-size factor is limited.Source: 1) Service Attributes · p. 23

2) Service Analysis:

  • Service analysis must consider spatial and temporal distributions because service attributes affect resource-block occupancy, interference, and processing performance.The reviewed approaches combine stochastic geometry with queueing theory to model spatial users, task arrivals, and dynamic queuing; many queueing studies assume infinite storage despite IoT storage limits.Source: 2) Service Analysis · pp. 23–24

E. Summary Remarks

The review covers JCC-SAGIN resource-management modeling and optimization, from performance metrics and offloading models to traditional and AI-based solution classes. For user-platform association, it describes integer-programming and game-theoretic approaches to coupled decisions.

E. Summary Remarks

  • JCC-SAGIN modeling covers task completion time, energy, weighted latency-energy, QoE, AoI, and computing reliability, alongside binary or partial offloading and network and service models.Offloading choice depends on whether tasks can be divided; the review notes that task-queue design for different service needs remains a future topic.Source: E. Summary Remarks · pp. 24–25

A. Traditional Optimization Methods in JCC-SAGIN

1) User-Platform Association:

  • User-platform association is binary, making offloading problems mixed-integer nonlinear programs when resource-limited IoT devices rely on nearby space or aerial platforms.Relaxation-based methods may round continuous solutions without guaranteeing local optimality, while exact methods can require repetitive linear relaxations and longer runtimes.Source: 1) User-Platform Association · p. 25
  • Potential games model interacting users’ offloading strategies and can yield Nash equilibria, offering distributed approaches when centralized optimization is NP-hard.The review also describes their use for resource-block allocation and UAV–UGV matching.Source: 1) User-Platform Association · p. 26

2) Power Allocation:

The review surveys traditional JCC-SAGIN optimization across power, communication, computation, mobility, queues, and coupled variables. It then introduces learning-based decision methods as an alternative to limitations of traditional optimization.

2) Power Allocation:

  • NTN power allocation must account for severe path and atmospheric loss, changing user–platform distances, and energy-constrained UAVs; interference-free SNR problems can use KKT conditions, while SINR problems use DC, SCA, or quadratic transforms.Other reviewed approaches include sequential fractional programming, bilevel programming, and first-order approximations for energy-efficiency or power optimization.Source: 2) Power Allocation · pp. 26–27

3) Precoding and Beamforming:

  • Precoding and beamforming research moves beyond single-antenna models toward MIMO methods for path loss and offloading, including QT-based fractional programming and weighted-MMSE transformations.The review also notes QT applications to multi-beam satellite beamforming.Source: 3) Precoding and Beamforming · p. 27

4) Task Assignment:

  • After user association, task assignment allocates processing across platforms, using convex optimization, SCA, or prospect-theory-based non-cooperative games in the reviewed settings.The cited examples include NLoS UAV-ground links and user heterogeneity or possible UAV MEC-server failure.Source: 4) Task Assignment · p. 27

5) Trajectory Planning:

  • UAV placement is addressed with exhaustive search or sequential unconstrained convex minimization, while SCA iteratively tightens bounds for trajectory optimization.The review states that SCA does not require the JCC problem to be convex.Source: 5) Trajectory Planning · p. 27

6) Task Queue Management:

  • Stochastic task arrivals and limited computing capacity create queues at IoT devices and SAPs, making queue stability and computation efficiency joint concerns.Lyapunov optimization decomposes long-term problems into real-time decisions using drift-plus-penalty minimization at each time slot.Source: 6) Task Queue Management · p. 27

7) Addressing Multi-Variable Coupling:

  • Because coupled JCC variables make direct problems NP-hard, block coordinate descent and alternating optimization solve subsets of variables in turn to improve tractability.For two variable blocks, BCD operates as alternating optimization and can progressively refine a solution.Source: 7) Addressing Multi-Variable Coupling · p. 27

B. Learning-Based Intelligent Decision-Making Methods in JCC-SAGIN

  • Learning-based methods are introduced as a way to avoid time-consuming optimization iterations by learning environmental features and policies; ML can predict dynamic changes, map complex factor relationships, and automate management.The review presents these methods as relevant to intelligent edge computing in SAGIN.Source: B. Learning-Based Intelligent… · pp. 27–28

1) Markov Decision Process:

The review describes MDP-based learning for resource decisions in dynamic SAGIN, including deep-learning, reinforcement-learning, and multi-agent approaches. It also identifies scaling, training, and centralized-coordination costs as constraints on these methods.

1) Markov Decision Process:

  • MDPs model computation offloading through states, actions, rewards, transition probabilities, and policies; extensions include multi-objective MDPs and POMDPs for partially observed systems.Examples include jointly optimizing latency, energy, and collected tasks, and maximizing utility in multi-UAV sensing and computation.Source: 1) Markov Decision Process · p. 28

2) Deep Learning:

  • Deep learning supports complex input-output mappings and applications such as association, allocation, routing, congestion control, and demand prediction; a GNN-A2C framework was reported to converge in an aerial Edge IoT simulation.A two-layer UAV MEC method combines differential evolution for UAV location with distributed DNNs for offloading.Source: 2) Deep Learning · p. 28

3) Reinforcement Learning:

  • RL uses sequential state observations and actions to refine policies, while DQN suits discrete actions and DDPG handles continuous variables; DRL applies learning to dynamic environments without prior knowledge.The review describes uses including offloading, trajectory optimization, and UAV control.Source: 3) Reinforcement Learning · pp. 28–30
  • Learning methods face resource-consuming DRL training and long convergence times, while centralized training and execution can add latency, energy use, and cost in SAGIN.The review calls for adaptive DTDE algorithms that account for platform or device features and deployed resources.Source: 3) Reinforcement Learning · pp. 29–30
  • Multi-agent methods include CTCE, CTDE, and DTDE; DTDE reduces reliance on centralized processing through message interaction among agents.CTDE trains centrally using global state but allows SAP agents to act independently from local observations.Source: 3) Reinforcement Learning · p. 29
  • Compared with MADDPG, MATD3 significantly outperforms in total system cost for UAV trajectory, offloading, and resource-allocation decisions in a dynamic MEC environment.The review motivates MATD3 by noting that finding optimal MADDPG actions becomes difficult as SAP numbers increase and performance deteriorates in high-dimensional continuous action spaces.Source: 3) Reinforcement Learning · p. 30

4) Federated Learning:

Federated learning offers local model training without sharing device data, but its use in dynamic, heterogeneous SAGIN raises coordination and mobility challenges. The review also summarizes transfer learning and the broader resource-management themes.

4) Federated Learning:

  • Federated learning keeps data on end devices while sharing model parameters, but centralized learning procedures can impose signaling, computation, delay, energy, and synchronization costs in large SAGIN.The review also notes that global information collection may be unsuitable for private tasks and that SAP mobility should inform FL framework design.Source: 4) Federated Learning · pp. 30, 32
  • Reported FL applications include improved prediction accuracy over centralized and separate learning schemes and lower UAV energy consumption, while SAP movement remains a challenge and satellite-network FL research is at an initial stage.The review also describes federated DRL with higher average reward and lower execution delay than baselines within limited iterations.Source: 4) Federated Learning · p. 31

5) Transfer Learning:

  • Transfer learning lets agents reuse prior knowledge from similar tasks, and reviewed aerial-edge work applies it to UAV information sharing, deployment, and resource allocation.Source: 5) Transfer Learning · p. 32

C. Summary Remarks

  • The review spans traditional and learning-based methods, highlighting user-platform association, power, beamforming, assignment, trajectory, queues, and coupled variables, alongside ML’s potential for robustness, complex mapping, and rapid response.It identifies advanced beamforming to mitigate NTN path loss, particularly for task offloading, as an area of interest.Source: C. Summary Remarks · p. 32

VII. FUTURE RESEARCH DIRECTIONS IN JCC-SAGIN

The survey identifies further JCC-SAGIN design needs across system levels, including coding and multiple access. It discusses coding-enhanced distributed computing and access methods as ways to address network and task demands.

VII. FUTURE RESEARCH DIRECTIONS IN JCC-SAGIN

  • Further JCC-SAGIN design is needed across system levels, including coding, multiple access, and security.Source: §VII · p. 32

A. Coding-Enhanced Distributed JCC-SAGIN

  • Coded storage-and-computation balances communication, caching, and computing, while coded distributed computing adds redundancy to mitigate failures and stragglers.Its task-partition ratio and redundant load require careful design; related UAV and satellite offloading work examines uncertainty and delay-energy tradeoffs.Source: A. Coding-Enhanced Distributed JCC-SAGIN · p. 33

1) Orthogonal Multiple Access:

  • OFDMA is more widely used than TDMA for delay-sensitive computation tasks, but co-channel interference can arise when orthogonal channels are insufficient or frequencies are reused.With orthogonal uplinks, interference can be omitted; multiple UAVs using the same subchannel can instead interfere.Source: 1) Orthogonal Multiple Access · p. 33

2) Non-Orthogonal Multiple Access:

  • NOMA lets multiple users offload computation simultaneously on the same time-frequency resource block using successive interference cancellation, but its performance depends on system design, channel conditions, and application.The survey describes potential gains in spectrum use, connectivity, adaptability, energy efficiency, and multi-task offloading, while calling for detailed performance evaluation.Source: 2) Non-Orthogonal Multiple Access · pp. 33–34

3) Combination of OMA and NOMA:

The survey reviews access-mode combinations, blockchain and quantum approaches, and maritime JCC applications, then recaps its broader review of JCC-SAGIN. It highlights open design tradeoffs and constraints across these areas.

3) Combination of OMA and NOMA:

  • Combining OMA and NOMA and designing their ratio according to task attributes offers an alternative to the usual binary access-mode choice.The modes have different advantage areas in simulation; the survey contrasts OMA's low complexity and lack of intra-cell co-channel interference with NOMA's spectral-efficiency gains, especially with multiple users.Source: 3) Combination of OMA and NOMA · p. 34

1) Blockchain:

  • Blockchain can provide decentralized SAGIN control, but sequential validation adds latency, chain replication consumes storage, and computing burdens IoT devices and SAPs.The surveyed architecture records data with immutable cryptographic signatures, hashes, and consensus mechanisms.Source: 1) Blockchain · p. 34

2) Quantum-Based Approaches:

  • Satellite QKD work reports a kilohertz key rate over up to 1200 kilometers, while integrated fiber and free-space QKD links extend communication distance up to 4600 kilometers.Source: 2) Quantum-Based Approaches · p. 35

D. Integration With Maritime Applications

  • Maritime JCC work includes hybrid satellite-UAV rate maximization with only position-related large-scale CSI and hierarchical link scheduling and rate adaptation to minimize energy while meeting QoS.The survey notes that computation-offloading work in UAV-assisted marine networks remains limited.Source: D. Integration With Maritime… · p. 36

VIII. CONCLUSION

  • The survey reviews JCC-SAGIN architecture, enabling technologies, applications, resource-management modeling and optimization, and prospective research directions.It frames this work against a gap in comprehensive research on NTN computing functions.Source: §VIII · p. 36
Loading 2403.17400v3…