PaperlayerView exact revision on arXiv

Source-linked AI summary

Towards Federated, Green, and Resilient 6G Non-Terrestrial Networks

Sarath Babu, Victor Baños-Gonzalez, Mario Cordina, Debabrata Dalai, Tomaso de Cola, Franco Davoli, Etienne Victor Depasquale, Ashutosh Dutta, Hesham ElBakoury, Michael A. Enright, Giovanni Giambene, Sumit Goswami, Ramesh Gupta, Wael Jaafar, Eman Hammad, B. S. Manoj, Tony Li, Manuel M. H. Roth, Paresh Saxena, Pat Scanlan, Zhili Sun, Daniele Tarchi, Saviour Zammit

arXiv:2609.05184v1cs.NI

TL;DR

The paper addresses how terrestrial and non-terrestrial networks can be federated despite interoperability, management, resilience, and security challenges. It combines architectural analysis, orbital-network simulation, Open-RAN considerations, and security mechanisms, finding that federation improves latency and connectivity robustness while supporting resilient service continuity.

  • Problem

    Integrating terrestrial and non-terrestrial networks requires interoperability, unified management, resource coordination, and resilience across heterogeneous and distributed domains.

  • Method

    The paper studies federated T-NTN architectures, Open-RAN function placement, edge intelligence, energy efficiency, and security using architectural analysis and orbital-network simulation.

  • Results

    Federation achieves up to 23% lower average delay, eliminates call blocking, and reduces worst-case latency by nearly 47%.

  • Takeaways & Limitations

    Federated, cloud-native, service-based T-NTN architectures support global coverage, service continuity, cross-domain resource management, and stronger resilience.

Abstract

from arXiv · show

This study focuses on future Non-Terrestrial Networks (NTN) integrated with Terrestrial Networks (TN) for future 5G/6G systems. NTN envisions a 3D architecture, where Low Earth Orbit (LEO) satellite networks will play a key role in bridging the digital divide, complementing the gradual terrestrial 5G/6G rollout concentrated in high-density and high-traffic areas, by ensuring service continuity across broad geographic regions and providing coverage in case of emergencies or in remote areas. In this context, we address networking issues for the integration and federation of Terrestrial and Non-Terrestrial Network (T-NTN) in line with the IMT-2030 vision, focusing on interoperability, spectrum coexistence, unified control and management, and service continuity. Federation is a complementary approach to integration that enables distinct satellite systems to cooperate through agreements, potentially unified satellite terminals, and common resource management. We show that system federation significantly enhances both latency performance and connectivity robustness compared with non-federated LEO architectures. This paper also investigates the challenges and possible solutions for adopting the Open-RAN architecture for T-NTN, including routing options for mega-LEO systems, edge intelligence, and energy efficiency as critical elements for sustainability. Finally, we address the security, privacy, and resilience aspects of federated T-NTN architectures with emphasis on zero-trust, secure routing, trustworthy edge intelligence, Post-Quantum Cryptography (PQC), and Quantum Key Distribution (QKD).

I. INTRODUCTION

NTNs are presented as a complement to terrestrial networks for extending connectivity, supporting resilient coverage, and enabling integrated 5G/6G services. The section frames federation, interoperability, and multi-orbit operation as responses to coverage, latency, routing, and resilience challenges.

  • NTNs can bridge connectivity gaps in remote areas and provide resilient communications in disaster-prone regions.
  • LEO mega-constellations have expanded access performance toward terrestrial mobile broadband while supporting broad-area coverage.Starlink’s median downlink bit rate has increased as its constellation expanded.
  • Direct-to-Device satellite links aim to connect conventional smartphones and IoT devices without specialized large-antenna terminals.The approach is intended to extend connectivity to underserved and remote areas.
  • Satellite connectivity still faces high-power and antenna requirements, limited compatible smartphones, and spectrum-regulatory challenges.Large electronically steerable or high-gain antennas are needed to close the link budget with standard handsets.
  • Multi-orbit systems trade coverage, capacity, and latency by combining persistent GEO coverage with lower-delay LEO and MEO operation.These architectures require routing schemes suited to their three-dimensional structure.
  • The INGR Satellite Working Group examines TN–NTN integration through roadmaps covering open and federated architectures, routing, mobility, and edge intelligence.
  • Federated multi-orbit architectures are motivated by the fragility of isolated mega-constellations under collision risks and severe space weather.Traffic can be redistributed across heterogeneous systems to improve resilience and service continuity.
  • Federation and interoperability support shared resources, roaming, service continuity, spectrum coexistence, and unified management across independent network domains.The paper places these capabilities within the IMT-2030 vision for integrated terrestrial and non-terrestrial networks.

III. SPECTRUM ALLOCATIONS

Spectrum allocation is a foundational constraint and enabler for implementing and expanding 5G and 6G networks. The section situates NTN spectrum within international allocation processes and presents reference tables for NTN band numbering and U.S. supplemental coverage bands.

  • Spectrum allocation remains a fundamental enabler and constraint for implementing and expanding 5G and 6G networks.
  • The ITU-R World Radio Conference identifies spectrum for IMT and designates FSS/MSS bands under which NTN can operate.
  • 3GPP defines TN and NTN air-interface requirements within the allocated spectrum, including band numbers and radio specifications.
  • Table 3 organizes NTN bands according to a numbering convention.
  • Table 4 lists U.S. spectrum bands from 600 MHz to 2 GHz designated for supplemental coverage from space.

A. MOBILE SATELLITE SERVICE (MSS) D2D SPECTRUM

The section examines multi-layer NTN architectures and O-RAN functional splits for integrating terrestrial and satellite networks. It emphasizes distributing processing across space and ground while balancing scalability, synchronization, power, computation, and service requirements.

  • Architecture vision: 6G-NTN envisions a multi-layer, multi-orbit 3D space system integrated into the broader 6G ecosystem.
  • O-RAN architecture: O-RAN disaggregates the gNB into independent Radio Unit, Distributed Unit, and Centralized Unit components connected through standardized interfaces.
  • Functional splits: Split 1 places a full gNB in space, terminating the 5G NR protocol stack and enabling space-based participation in handover procedures.
  • Design constraints: Space-ground CU-DU deployment requires persistent connectivity, while in-space CU-DU signaling must fit within satellite power and computational budgets.
  • Functional splits: Split 2 separates CU and DU functions across NTN nodes, with their F1 interface carried over inter-satellite links when both are deployed in space.
  • Design constraints: Split 7 keeps the RU in space but requires tight RU-DU synchronization, and RIC placement depends on the adopted functional split.

V. INTEROPERABILITY AND FEDERATION OF NTN

Federation allows independent satellite constellations to cooperate by sharing routing and onboard computational resources. This can provide alternative paths and load balancing, while integrated TN-NTN systems must also address shared spectrum and handover paths.

  • Federation enables two independent constellations to share routing and onboard computational resources rather than operating as isolated systems.
  • Shared resources can offer potential latency improvements through load balancing, alternative routing, and reduced queuing.
  • Federated satellite and terrestrial networks must coordinate spectrum use and handover paths between domains.

A. SYSTEM MODEL FEDERATION IN NTN SYSTEMS

The system model compares end-to-end latency in non-federated and federated satellite routing. It represents time-varying links, propagation and transmission delays, operator boundaries, and federation overhead to evaluate shorter feasible paths.

  • The model compares a single-constellation case with a cooperative two-constellation case using end-to-end user latency.
  • 70-71: The analytical model focuses on propagation and transmission delays and examines longer non-federated paths versus shorter federated alternatives.The model ignores queuing and processing overheads.
  • Each graph edge combines propagation and transmission delay, while an availability indicator captures time-varying visibility and link outages.
  • Federation overhead represents authentication, gateway translation, or routing-policy costs associated with using partner-operator edges.
  • Federated routing expands the feasible path set beyond the non-federated operator-specific paths, subject to operational edge availability.
  • Federation may reduce latency through shorter paths or fewer hops, but its effectiveness depends on topology, satellite availability, and inter-operator overheads.
  • The latency difference is interpreted as positive when federation reduces latency and non-positive when federation increases it.

B. SIMULATION SETUP FOR FEDERATION IN NTN SYSTEMS

The simulation models federated and non-federated Iridium Next and Starlink architectures over orbital and terrestrial networks. It uses satellite visibility, topology, and routing computations to evaluate performance under different source-destination conditions.

  • Simulation framework: The simulation uses orbital propagation and visibility analysis with Skyfield and validates constellation visualizations with STK.
  • Simulation framework: Network topology construction and path computation are implemented with Python libraries, including NetworkX.
  • Architectures: Non-federated Iridium routing permits only Iridium satellites and prohibits inter-constellation communication.
  • Architectures: Non-federated Starlink routing is restricted to Starlink satellites and excludes connectivity to Iridium Next.
  • Architectures: Federated routing enables intra- and inter-constellation ISLs when satellites from different constellations are within communication range.
  • Topology and traffic: The evaluated topology includes 75 Iridium satellites, 121 polar Starlink satellites, and globally distributed ground stations serving 100 UEs each.
  • Topology and traffic: Two source UE-destination UE selection strategies represent average-case and worst-case conditions for evaluating Dijkstra routing performance.

1) Shortest Path – Random UE

The evaluation compares shortest-path routing under random and farthest-user selection across non-federated and federated LEO architectures. Federation reduces delay, hop counts, variability, and call blocking, while farthest-user routing increases delay but can produce more deterministic paths.

  • Routing scenarios: SP-RU selects source and destination UEs uniformly, whereas SP-FU selects the geographically farthest UE pair for shortest-path evaluation.Both strategies compute paths with Dijkstra’s algorithm; SP-RU represents average-case traffic, while SP-FU represents a worst-case spatial condition.
  • Latency: 68.44 ms is the federated architecture’s lowest median SP-RU E2E delay, compared with 73.44 ms for Starlink and 88.94 ms for Iridium Next.Federation reduces average E2E delay by approximately 23% versus non-federated Iridium Next and 6.8% versus non-federated Starlink.
  • Path efficiency: 3.79 average hops is the federated SP-RU result, versus 5.29 for Iridium Next and 4.15 for Starlink; under SP-FU, the values rise to 5.97, 7.51, and 6.28, respectively.The federated architecture also has narrower routing-path distributions, indicating more consistent paths across user locations.
  • Connectivity: 0% call blocking is achieved by federation under both SP-RU and SP-FU, while Iridium Next records 14.2% and 12.2%, and Starlink remains below 5%.Call blocking denotes the absence of a feasible inter-satellite routing path at call initiation.
  • Latency distribution: 47% is the nearly reduced SP-RU worst-case latency, falling from 231.24 ms for non-federated Iridium Next to 122.22 ms for federated Iridium-Starlink.The SP-RU delay CDF shifts left for the federated architecture, indicating lower overall delay.
  • Latency stability: 0.08 is the federated SP-FU latency CV, compared with 0.21 for Iridium Next, showing lower variability despite higher average delay under farthest-user routing.Under SP-RU, federation reduces variability by approximately 37% versus non-federated Iridium Next, from CV 0.43 to 0.27.

D. OPEN RESEARCH PROBLEM FOR FEDERATION IN NTN SYSTEMS

Federating NTN systems requires standardized interfaces, interoperable network functions, coordinated operator procedures, and routing that adapts to dynamic, heterogeneous constellations. The research agenda also spans resilience, AI-assisted control, and scalable management across orbital layers.

  • Federation architecture: The federation interface should standardize monitoring, resource discovery, QoS-based negotiation, reservation, lifecycle management, and security between operators.A standardized API such as RESTful API is identified as one possible exposure mechanism.
  • Interoperability: Inter-operator federation requires roaming and handover protocols, authentication and trust frameworks, and Service Level Agreement coordination.Interoperability can also use unified control-plane APIs, common network-function abstractions, and standardized service exposure.
  • Requirements: Future NTN routing must support congestion-aware broadband, seamless TN–NTN handovers, post-quantum security, and resilience against DoS attacks, jamming, and node or link failures.Path diversity and backup alternatives are identified as mechanisms for robust routing.
  • Routing challenges: Routing must handle predictable but frequent handovers while limiting signaling overhead and centralized-control delays across large ISL networks.Segment Routing, SDN, backup paths, hierarchical designs, and distributed domains are identified as relevant approaches.
  • Edge intelligence: Limited onboard processing means ground-based network management remains essential in most current satellite-network designs.AI-native approaches include LSTM and transformer traffic forecasting, GNN topology modeling, and DRL for load balancing and fault recovery.
  • Multi-orbit federation: Federated and multi-orbit systems need paths selected across orbital layers according to latency, reliability, and resource availability.Hierarchical control coordinated by ground stations and coupled with AI-driven predictive analytics are proposed directions.

A. INSIGHTS FROM SYSTEM-LEVEL SIMULATIONS

System-level simulations compare centralized shortest-path, ideal global-information, and distributed load-balancing routing under different QoS priorities. Centralized approaches minimize latency for delay-critical traffic, while distributed routing offers scalable load balancing for relaxed-latency services.

  • Simulation setup: The simulator evaluates QoS compliance using end-to-end latency and packet-dropping rate in a reference mega-LEO routing scenario.Average session latency is compared for Shortest Path First, an ideal “God’s Eye View” benchmark, and distributed load-balancing routing.
  • Delay-critical QoS: Centralized Shortest Path First and God’s Eye View routing achieve lower latency than distributed routing for delay-critical QoS classes.Distributed local decisions can create longer zigzag paths that occasionally exceed the 150 ms latency constraint.
  • Routing trade-offs: The simulations highlight a trade-off between centralized informed decisions for latency and distributed localized decisions for scalability and load balancing.The results emphasize the value of rapid, informed routing decisions in NTN systems.
  • Throughput: Distributed routing remains attractive for relaxed-latency services because load balancing can provide higher throughput than Shortest Path First.The throughput comparison relates observed throughput to offered network load.
  • Best-effort QoS: Drop rates below 10^-6 are achieved by the distributed scheme for best-effort service, whereas Shortest Path First is prone to congestion and high packet loss.The ideal God’s-Eye-View benchmark can completely avoid packet drops because it has immediate complete network knowledge.

VII. EDGE INTELLIGENCE

Edge Intelligence brings local AI processing and distributed learning to T-NTNs, while O-RAN, network slicing, and energy-aware management support deployment. Major constraints include resource limitations, model issues, security risks, and difficulty mapping virtualized functions to physical energy states.

  • Edge Intelligence: Edge Intelligence processes data and makes decisions near its source, reducing propagation latency and improving bandwidth efficiency in T-NTNs.Satellites and aerial platforms can process tasks locally or serve as distributed offloading platforms.
  • Architectural enablers: O-RAN provides a virtualized framework for deploying intelligent RAN functions across distributed T-NTN infrastructure, while network slicing supports customized services.
  • Distributed learning: Federated Learning exchanges model updates instead of raw data, enabling collaborative training with privacy preservation across resource-constrained NTN nodes.Distributed learning also includes Split Learning, which partitions models between devices and other processing locations.
  • Challenges: Dynamic node conditions make learning-function placement and resource allocation complex, while data heterogeneity, model drift, and distributed security vulnerabilities can reduce performance.
  • Energy management: GAL introduces Energy Aware States and abstraction interfaces, but virtualized multi-tenant environments make direct mapping from virtual resources to hardware energy states difficult.Shared hardware, hypervisors, orchestration platforms, and virtual resources complicate energy attribution.
  • Energy efficiency: Energy evaluation should include both computation and communication, including the differing loads of full-BS versus DU/CU-split deployments.Near-Real-Time RIC placement and dynamic xApp or dApp placement involve responsiveness–energy trade-offs.
  • Energy efficiency: In the YOLOv8n workload, GPU and NPU offloading reduced AI processing power by approximately 15 W relative to CPU execution.CPU power consumption increased by approximately 20 W during execution.

IX. SECURITY, PRIVACY, AND RESILIENCE IN FEDERATED

Federated T-NTNs expand security requirements across operators, domains, and intelligent edge components. The proposed direction combines trust management, secure routing, privacy-aware learning, and joint optimization of performance, energy, and security, while leaving interoperability and long-term cryptographic management open.

  • Threat landscape: Integrated TN-NTNs expand the attack surface across ground, aerial, multi-orbit, control-link, gateway, inter-satellite, and edge-computing domains.
  • Trust and security: Federation requires lifecycle-managed identities, end-to-end protection, post-quantum-ready key exchange, and confidential cross-domain security telemetry.
  • Secure routing: Federated routing can incorporate latency, capacity, risk scores, trust levels, and regulatory constraints into link weights and feasible-path selection.
  • Trustworthy edge intelligence: Federated and distributed learning can exchange model updates and gradients rather than raw data, but remains vulnerable to poisoning, inference attacks, and backdoored models.
  • Security–energy trade-offs: Aggressive energy saving can reduce monitoring or cryptographic protection, whereas strong security controls impose energy and bandwidth overheads.
  • Open issues: Secure federated T-NTNs still require interoperable security policies and SLAs, privacy-preserving telemetry sharing, resilient AI defenses, and crypto-agility for long-lived satellites.

A. EXPERIMENTAL RESULTS: CLOSED-LOOP AUTOMATION

The closed-loop framework detects and mitigates control- and user-plane DoS attacks while dynamically orchestrating resources. Testbed results show recovery of service quality without interrupting ongoing communication.

  • Framework: The framework combines real-time monitoring, intrusion detection, SDN, and NFV for automated anomaly response.
  • Framework: Mitigation and service provisioning run in parallel, allowing additional VNFs and resources to compensate for shortages while enforcing security policies.
  • Detection and enforcement: The intrusion-detection system analyzes control- and user-plane traffic, then sends alerts to an SDN controller that configures enforcement elements.
  • Experimental setup: The testbed emulates realistic 5G conditions using commercial and open-source tools for traffic generation, attack simulation, service provisioning, monitoring, and detection.
  • Attack scenarios: Three DoS scenarios target AMF signaling, UPF bandwidth, and related network resources, with orchestration dynamically allocating additional capacity.
  • Results: Packet loss decreases and voice MOS returns to acceptable levels after mitigation, without interrupting ongoing communication.
  • Implications: Closed-loop detection, mitigation, and orchestration support service continuity while network resources are managed during attacks.

X. CLOUD-NATIVE ARTIFICIAL INTELLIGENCE-BASED

Cloud-native architectures are presented as a way to scale secure 6G data and AI/ML processing amid growing computing demand and cybersecurity threats. The described designs combine streaming and batch pipelines with agentic orchestration and external tools.

  • Cybersecurity context: Zero Trust is positioned as a cybersecurity approach for hardening increasingly connected infrastructure.
  • Cloud-native motivation: Cloud-native design uses microservices, containers, orchestration, and APIs to provision scalable and agile cloud resources.
  • Cloud-native motivation: Growing demand for computing resources, especially from GenAI workloads, and data-center power concerns motivate alternative infrastructure directions.
  • Architecture: A secure 6G cybersecurity architecture should process large streaming and batch datasets and integrate AI/ML capabilities.
  • Data processing: The data-processing architecture integrates multiple sources with batch and stream pipelines that clean and distribute incoming data.
  • AI/ML processing: The AI/ML architecture uses an agentic orchestrator to manage cloud task agents and connect them to external agents and tools through MCP.
  • Conclusion: Cloud-native design creates opportunities for both data-processing and AI/ML cybersecurity architectures.

XI. QUANTUM SECURITY ARCHITECTURE AND INTEROPERABILITY

Federated T-NTN architectures require coordinated architectural, operational, and security measures, including zero-trust protection and quantum-resilient key management. The paper positions PQC and QKD as complementary responses while emphasizing interoperable, federated infrastructure.

  • Quantum security: Quantum computers threaten current encryption, motivating PQC for near-term protection and QKD for physically grounded secure key exchange.PQC uses classical algorithms designed to resist quantum attacks, while QKD enables key exchange based on quantum-mechanical principles.
  • Quantum security: Satellite quantum communication has been demonstrated over more than 1,200 km, supporting the feasibility of satellite-based quantum networks.The Micius mission demonstrated satellite-based quantum entanglement distribution between ground stations in 2017.
  • Interoperability: Quantum and classical optical communications can be integrated through coexistence approaches to support interoperability.The passage identifies coexistence as one of two main integration methods for quantum and classical optical communication.
  • Architecture and operations: Federated T-NTN designs improve routing efficiency and resilience while supporting global coverage, service continuity, and heterogeneous terrestrial, aerial, and space deployments.The reported improvements include up to 23% lower average delay, elimination of call blocking, and nearly 47% lower worst-case latency.
  • Architecture and operations: O-RAN supports federated T-NTN operations through edge intelligence, dynamic function placement, resource optimization, and network slicing.Its hierarchical RIC ecosystem spans Non-RT and Near-RT functions across terrestrial and non-terrestrial segments.
  • Security architecture: Zero-trust federation continuously authenticates, authorizes, and verifies satellites, gateways, edge nodes, and administrative domains.The architecture also requires strong identity management, end-to-end protection, secure cross-domain policy exchange, and AI-driven closed-loop mitigation.
Loading 2609.05184v1…