Physics-Based Trajectory Design for Cellular-Connected UAV in Rainy Environments Based on Deep Reinforcement Learning

Hao Qin, Zhaozhou Wu, Yuanwei Liu, Xingqi Zhang, Xinyue ZhangIEEE Trans. ITS 2025 · DOI

TL;DR

This paper tackles the challenge of designing optimal flight paths for drones communicating via cellular networks in rainy weather. By combining a physics-based simulator for rain's impact on radio signals with a deep reinforcement learning approach, they enable drones to navigate effectively and maintain reliable communication even in adverse conditions.

What Problem Does This Paper Address?

Unmanned Aerial Vehicles (UAVs), or drones, are increasingly being integrated into cellular networks to extend their range and communication capabilities. However, this integration faces significant hurdles, particularly in challenging weather like rain. Rain causes signal attenuation, weakening the communication link between the drone and ground base stations. Furthermore, existing cellular infrastructure is primarily designed for ground users, leading to suboptimal coverage in the sky. Traditional trajectory optimization methods struggle with the complexity of these real-world conditions, often relying on time-consuming measurements or simplified models. This paper aims to bridge this gap by developing a robust trajectory design method that accounts for the physics of radio wave propagation in rainy environments.

Methods

The core of this work lies in a two-pronged approach:

  • Physics-Based Wave Propagation Simulation: To accurately model the impact of rain on radio signals, the authors employ the Parabolic Wave Equation (PWE) method. This deterministic approach considers detailed environmental geometry and the physical effects of raindrops on wave propagation, including attenuation and diffraction. This replaces the need for extensive, costly real-world measurements.
  • Deep Reinforcement Learning (DRL) for Trajectory Optimization: The trajectory design problem is formulated as a Markov Decision Process (MDP). A Dueling Deep Q-Network (DDQN) with multi-step learning is then used to train the UAV's policy. This DRL approach allows the UAV to learn optimal flight paths by balancing two key objectives: minimizing flight time and maximizing the Signal-to-Interference-plus-Noise Ratio (SIR) along its path. The multi-step learning and dueling architecture are incorporated to improve training stability and efficiency.

Datasets

Not discussed in the paper. The authors generate their own "datasets" through simulations using the PWE simulator and the DRL framework.

Key Insights and Ideas

  • Rain's Significant Impact: The paper clearly demonstrates that rain has a substantial impact on radio signal strength and, consequently, on optimal UAV trajectories. Ignoring this can lead to suboptimal or failed missions.
  • Physics-Based Simulation is Crucial: Relying solely on empirical models or ignoring environmental factors like rain is insufficient for robust UAV communication. A physics-based approach, like PWE, is necessary for accurate modeling.
  • DRL for Complex Trade-offs: Deep Reinforcement Learning, particularly with enhancements like dueling networks and multi-step learning, is well-suited to handle the complex trade-off between minimizing flight time and maximizing communication quality in dynamic environments.
  • Adaptive Trajectories: The optimal trajectories differ significantly between homogeneous atmospheric conditions and rainy conditions, highlighting the need for adaptive path planning.

Key Results

  • The proposed physics-based DRL approach successfully designs UAV trajectories that avoid weak coverage regions and reach the destination.
  • The PWE simulator's results for rain attenuation show good agreement with the ITU-R model, validating its accuracy.
  • The study demonstrates that the weight assigned to flight time versus SIR in the objective function directly influences the resulting trajectory, allowing for customization based on mission priorities.
  • The algorithm effectively handles varying rain rates, showcasing its robustness.
  • The paper also briefly explores the impact of wind conditions on trajectories, suggesting further avenues for research.

Limitations and Future Work

  • Fixed Altitude Assumption: The current model assumes a fixed flight altitude for the UAV. Future work could explore 3D trajectory optimization.
  • Simplified Environment: While rain is considered, other complex environmental factors like fog, snow, or dynamic obstacles are not explicitly modeled.
  • Computational Complexity: While improvements are made, DRL training can still be computationally intensive. Further optimization of the training process could be beneficial.
  • Wind Modeling: The paper introduces wind as a factor but suggests more in-depth investigation into its impact on wave propagation and trajectory design.
  • Real-World Validation: The results are based on simulations. Future work should include experimental validation in real-world scenarios.

Summary updated

BibTeX
@article{qin2025physics,
  author = {Hao Qin and Zhaozhou Wu and Yuanwei Liu and Xingqi Zhang and Xinyue Zhang},
  title = {{Physics-Based Trajectory Design for Cellular-Connected UAV in Rainy Environments Based on Deep Reinforcement Learning}},
  journal = {IEEE Trans. Intell. Transp. Syst.},
  year = {2025},
  doi = {10.1109/tits.2025.3550348},
}
Figure