TITAN: Twin-Informed Topology Adaptation for LAWN-enabled D2C Communication

TL;DR

Titan, a twin-informed topology adaptation framework that builds a high-fidelity Digital Twin of the affected urban area and performs site-specific, ray-traced air-to-ground channel modeling via Sionna RT informs a Bayesian optimization process that adapts the aerial topology to maximize coverage and Quality of Service (QoS) for ground users.

Abstract

Low-Altitude Wireless Networks (LAWN) are transforming the low-altitude airspace into a mission-driven, dynamically reconfigurable 3D network fabric for safety-critical and public-safety operations. In parallel, Direct-to-Cell (D2C) satellite access can rapidly restore connectivity after disasters, yet dense urban blockages make the satellite-to-ground link unreliable for many users. To overcome this, we leverage the LAWN aerial layer and form an adaptive low-altitude relay topology where Unmanned Aerial Vehicles (UAVs) act as D2C-assisted aerial relays for obstructed ground users. We introduce TITAN, a twin-informed topology adaptation framework that builds a high-fidelity Digital Twin (DT) of the affected urban area and performs site-specific, ray-traced air-to-ground channel modeling via Sionna RT. This informs a Bayesian optimization process that adapts the aerial topology to maximize coverage and Quality of Service (QoS) for ground users by using UAVs as optimal D2C relays. Extensive system-level simulations with Sionna show that TITAN consistently outperforms the baselines and delivers +32.2% user coverage, +64.9% system sum-rate, and +49.3% fairness over the state-of-the-art (SOTA) that employ heuristic placement or statistical channel approximations. To support further research in resilient network design, we open-source the codebase of the TITAN framework.

Figures (9)

• Figure 1: Proposed ray-traced Low-Altitude Wireless Network (LAWN)-enabled Digital Twin (DT) framework highlighting the core contributions: The DT ingests environment mesh and UE locations to compute precise channel impulse responses (C1), derives optimal 3D UAV placements via Bayesian optimization (C2) to maximize coverage/QoS, and enables reactive topology adaptation (C3) to environmental changes. Finally, it performs system-level simulation (C4) and data collection (C5).
• Figure 2: D2C signals directly reaching UEs are frequently blocked or severely attenuated. Instead, UAVs leverage stronger, more reliable D2C satellite links due to superior antenna alignment and guaranteed LoS, serving as aerial relays.
• Figure 3: The TITAN Framework Workflow and Simulation Example. The San Francisco mesh includes detailed building models and ground elevation variations to capture realistic urban topography. UEs are positioned 1.5 meters above the local ground surface to represent typical handheld device usage, buildings are modeled as blocks with marble and metal (roof) material types. After obtaining the UE GPS data and gNB information, TITAN creates Digital Twin of the environment. After gNB numbers 1 and 2 fail, TITAN analyzes the environment status and deploys LAWN response. Color bar represents the maximum path gain.
• Figure 4: Starlink satellite visibility analysis over San Francisco for 24-hour window using TLE data with elevation thresholds.
• Figure 5: Performance comparison of UAV placement methods in terms of sum-rate and coverage ratio. Sum-rate and coverage achieved by different UAV placement strategies as a function of the number of UAVs, demonstrating that the proposed TITAN optimization consistently outperforms state-of-the-art multidt2025 (+32.2% user coverage, +64.9% sum-rate) and other baselines.