Non-Terrestrial Networks: Architecture & Applications

• Non-Terrestrial Networks (NTN) are integrated communication systems combining satellites, high-altitude platforms, and terrestrial networks to provide ubiquitous, high-capacity connectivity.
• NTN architectures leverage diverse platforms—GEO, MEO, LEO, and UAVs—with tailored physical-layer adaptations to balance latency, throughput, and coverage.
• Research in NTN emphasizes dynamic traffic offloading, AI-driven resource management, and protocol enhancements to support IoT, disaster recovery, and high-speed broadband.

Non-Terrestrial Networks (NTN) constitute an integrated communications paradigm wherein spaceborne (LEO, MEO, GEO satellites) and airborne platforms (HAPs, LAPs, UAVs) are embedded into the global wireless landscape alongside terrestrial networks. By exploiting the multi-altitude physical layer, advanced RAN architectures, and adaptations to the 3GPP NR stack, NTNs provide wide-area, high-availability, and resilient connectivity in support of 5G/5G-Advanced and future 6G systems. Their integration is crucial for achieving ubiquitous broadband access, massive IoT coverage, and robust public-safety support, particularly across rural, maritime, aeronautical, and disaster-recovery domains (Hernandez et al., 2023).

1. NTN Architectures, Platform Classes, and System Topologies

NTN deployments are characterized by the diversity of their platform classes and the resulting architectural flexibility:

• GEO satellites ($h\approx35\,786$ km) provide global coverage with VHTS payloads delivering up to $100$--$300$ Gb/s per satellite in Ka/Ku bands, at a cost of high one-way delays ($\sim$550 ms) and negligible Doppler (Hernandez et al., 2023).
• MEO constellations ($7\,000$--$25\,000$ km) reduce round-trip latency to 150–200 ms; beams use multi-spot phased arrays and frequency reuse factors of $3$–$4$.
• LEO mega-constellations ($300$–$2\,000$ km; Starlink, OneWeb) achieve slant-range RTTs of $20$–$50$ ms, but face severe Doppler ($\sim$7.5 km/s satellite speed) and require frequent handover and spot-beam management. Typical LEO satellites offer per-unit capacities in the tens of Gb/s range.
• High-Altitude Platforms (HAPs, 18$-$22 km) and Low Altitude Platforms (LAPs, <1 km) (balloons, UAVs) offer near-terrestrial RTTs ($<$10 ms) and play a pivotal role in low-latency MBB and emergency restoration scenarios.

NTN–terrestrial integration, termed Space-Terrestrial Integrated Network (STIN), typically follows four topological templates:

• Star (UE$\rightarrow$SAT$\rightarrow$gateway$\rightarrow$5G Core)
• Satellite mesh (via ISLs for multi-hop LEO routing)
• Multi-homed gateways for redundancy
• Integrated Access/Backhaul (IAB), where NR-SAT nodes relay traffic as wireless backhaul (Hernandez et al., 2023).

2. Radio Propagation, Link Budget, and Physical Layer Adaptations

NTN link design is dominated by free-space path loss (FSPL), atmospheric absorption, and weather-driven attenuation. The path loss is:

$\text{FSPL (dB)} = 20 \log_{10}(4 \pi d / \lambda)$

where $d$ is the slant range, $\lambda = c/f$ is the wavelength. For $f=20$ GHz and $d=1\,000$ km, FSPL $\approx207$ dB (Hernandez et al., 2023). Atmospheric loss, $L_\text{atm}$, aggregates gaseous (ITU-R P.676), rain/cloud (Crane, ITU-R P.838), and scintillation (ITU-R P.618) effects. In Ka-band, heavy rain ($R=25$ mm/h) yields $>10$ dB attenuation; scintillation margins are typically $1$--$3$ dB.

Link budget in dB:

$$P_\text{rx} = P_\text{tx} + G_\text{tx} - L_\text{tx} - \text{FSPL} - L_\text{atm} - L_\text{add} + G_\text{rx} - L_\text{rx}$$

EIRP, antenna gains ($0$ dBi handset $\rightarrow$ $35$ dBi VSAT), system noise figures ($3$--$9$ dB), and bandwidths ($10$--$1\,000$ MHz per beam) are tightly specified to balance SNR, capacity, and availability under fading and interference (Hernandez et al., 2023).

NTN channels demand PHY/MAC waveform modifications:

• Cyclic prefix (CP) extension to accommodate large propagation delays
• Doppler compensation using GNSS-based frequency tracking per UE; subcarrier spacing adaptively increased ($15$ kHz $\rightarrow$ $240$ kHz) in LEO
• Frequency allocations: S/L bands for NB-IoT; C/Ku/Ka for broadband; Q/V/W for ISLs (Hernandez et al., 2023)

3. 5G NR–NTN Integration: Protocol Stack and Standardization

The 3GPP NR–NTN architecture integrates NR-SAT nodes within the NG-RAN, acting as gNBs (CU/DU) with feeder gateways interconnecting to the 5G Core via S1/Xn interfaces. NTN–specific protocol enhancements include:

• Random Access: Extended RAR windows (up to $10$ s for GEO, $5$ s for LEO), long PRACH preambles
• Timing Advance: GNSS-assisted slant-range measurement
• HARQ: Extended timers (two-way GEO $\sim811$ ms), augmented buffer management
• Scheduling: Quasi-static bearer for IoT, semi-persistent for MBB
• Link Adaptation: Larger TTI sizes (up to $2$ ms), static grants (Hernandez et al., 2023)

Doppler and handover across spot beams (LEO handover rates $>2$ Hz/UE) are addressed by broadcasted beam IDs, make-before-break dual-connectivity, and multi-satellite association for diversity.

Standardization is evolving through 3GPP Releases 18–20: support for NTN-specific bands (L, Ka), regenerative payloads and ISLs, enhanced IoT (RedCap UE, NB-IoT voice), GNSS resilience, and harmonized TN/NTN waveform and mobility management (Figaro et al., 8 Jan 2026).

4. Resource Management, Traffic Offloading, and Energy Efficiency

NTN–enabled STINs provide dynamic offloading and traffic shaping capabilities that are critical for energy and resource efficiency, especially under variable load:

• Dynamic association and power control: BLASTER (Block-Coordinate Gradient Ascent) jointly optimizes UE–BS association, BS transmit power, and bandwidth split between TN and NTNs (variable $\varepsilon$). Satellite-aided offloading leads to up to $45\%$ reduction in terrestrial energy consumption and $250\%$ throughput gains in rural scenarios, compared to 3GPP TN-only integration (Alam et al., 2024).
• Proportionally-fair optimization: Joint spectrum, association, and power allocation co-design yields mean rate increases $>200\%$ and nearly eliminates coverage holes, compared to static resource splits (Alam et al., 2023).
• Coverage extension: Any nonzero NTN bandwidth share reduces out-of-coverage UEs from $7\%$ to below $0.4\%$ in rural deployments (Alam et al., 2023).
• IoT traffic offloading: LPWAN (LoRa, SigFox, NB-IoT) adaptation to NTN uplinks, with optimal SF/rate/power allocation, dramatically improves capacity, offloading up to $30\%$ of the traffic with success-probability gains in dense scenarios (Wang et al., 2022).

5. Multi-Tier NTN in IoT, Disaster Recovery, and Content Broadcast

NTNs support a rich set of verticals and operational paradigms:

• Disaster communications: Multi-tier topology (Ground UEs$\to$UAV-BS$\to$HAPS-SMBS) and layered clustering algorithms (DLC-AHN) halve UAV backhaul energy under connectivity and QoS constraints, prolonging network lifespan during outages (Ozturk et al., 2024).
• Massive IoT: Three-dimensional gateway layering (UAV$\to$HAP$\to$LEO) optimizes both goodput and success probability; “LoRa+” SF randomization halves collisions and boosts capacity by $50\%$ (Wang et al., 2022). Coverage radii scale as TG $< 15$ km, UAV $< 8$ km, HAP $\sim100$ km, LEO $>1\,400$ km.
• Edge content broadcast: LEO-based NTN broadcast accelerates edge cache content placement by $20\%$–$33\%$ over standard TN, especially at high popularity skew and regional correlation (Wang et al., 2023).
• Hybrid backhaul: OGS$\to$FSO$\to$HAP$\to$IRS$\to$user chain, modeled via Gamma-Gamma and Nakagami fading, uses asymptotic outage/BEP/capacity formulas. Design trade-offs favor heterodyne over IM/DD detection, IRS proximity, and optimal relay gain configurations for SNR maximization (Shang et al., 3 Nov 2025).

6. System-Level Metrics, Stochastic Geometry, and Application Scenarios

Stochastic geometry (SG) characterizes platform spatial distributions, link reliability, and system capacity for NTN:

• Low-Altitude Platforms (LAPs): Modeled as planar PPPs ($\lambda_\text{LAP}$).
• HAPs/LEO: Binomial Point Process (BPP) or Cox process on spheres, mapping elevation, coverage, and LoS probability to closed-form connectivity and k-coverage metrics (Huang et al., 2023).
• Key metrics: Coverage probability $P_c(\theta)$, average rate, spatial throughput, dual-hop/relay availability, latency, and energy efficiency.
• Trade-offs: Higher altitude increases reach but degrades energy efficiency and capacity density; hybrid LAP+HAP+LEO deployments optimize the cost–coverage–latency continuum.

NTNs enable: (a) rural/remote broadband (criterion: area-availability $>99\%$ under fading margins), (b) disaster/post-catastrophe rapid deployment, (c) military/multiple-k-coverage for redundancy, and (d) global IoT backhaul (Huang et al., 2023, Jiang et al., 2023).

7. Challenges, Open Issues, and Future Research

Critical research fronts for NTN, all explicitly raised in the literature, include (Hernandez et al., 2023, Figaro et al., 8 Jan 2026, Alam et al., 2024):

• Doppler and protocol resilience: Robust PHY/MAC adaptation for maximal Doppler ($5$–$50$ kHz), especially in LEO
• Ultra-fast handover and mobility: Satellite diversity, dual-connectivity for multi-spot-beam management
• Resource slicing and orchestration: Algorithmic, AI/ML-driven solutions for joint TN–NTN resource allocation and dynamic slice isolation
• AI/ML integration: Predictive beam management, traffic/load forecasting, impairment mitigation, and cross-layer optimization
• 6G integration: RIS-enhanced links, quantum key ISLs, OAM multiplexing, advanced Layer-2 protocols for URLLC over GEO and HAP–ground multi-hop
• Standardization: 3GPP-NTN protocol evolution, onboard regenerative payloads, satellite edge computing, and harmonized spectrum definitions.

NTNs are thus formally established as a core enabler for next-generation wireless networks—realizing ubiquitous, resilient, and high-capacity connectivity by synchronously engineering architecture, physical-layer protocols, resource management, and AI-driven operations across integrated terrestrial, aerial, and orbital domains (Hernandez et al., 2023).