Abstract
This paper presents a novel 3D trajectory optimization framework for multiple drone cells (DCs) aimed at enhancing data collection from terrestrial IoT devices. By formulating the problem as a mixed-integer non-linear programming (MINLP) task and employing a Block Coordinate Descent (BCD) method, the authors jointly optimize user association, scheduling, and full 3D flight paths. Simulation results demonstrate that dynamic 3D trajectories significantly outperform static deployments, reducing average pathloss by 10-15 dB and improving user fairness by over 50%.
Introduction
The rapid expansion of the Internet of Things (IoT) faces significant connectivity challenges in urban and remote areas due to signal blockages and spectrum scarcity. While drone cells (DCs) provide a flexible solution for line-of-sight data collection, existing models often oversimplify drone movement to 2D planes or ignore backhaul constraints. This research addresses these gaps by designing optimized 3D trajectories that account for both user-to-drone (U2D) and drone-to-base-station (D2B) link qualities.
Methodology
The proposed approach decomposes the complex MINLP problem into four sub-problems: user association, scheduling, horizontal trajectory, and flying altitude. These are solved iteratively using Block Coordinate Descent (BCD). The horizontal path is optimized as a quadratic convex problem, while the altitude is determined through a quasi-convex optimization using the Newton-Raphson method.
Key Insights
The optimized paths reveal a "hovering effect," where drones spend the majority of their time directly above IoT devices to minimize pathloss. Furthermore, the D2B backhaul link acts as a physical floor, maintaining a minimum altitude (approximately 78 meters in simulations) to ensure reliable data offloading to the terrestrial base station. This mobility ensures that even edge users receive high-quality service, drastically reducing the variance in signal quality compared to static setups.
Results and Discussion
The 3D trajectory design achieves a substantial reduction in average U2D pathloss compared to static deployment benchmarks like the DI-PSO algorithm.
Quantitatively, the proposed method reduces pathloss by 10 to 15 dB. Additionally, the standard deviation of pathloss is reduced by an average of 56.66%. For instance, in a 7-drone scenario, the standard deviation dropped from 3.05 dB with static deployment to just 1.23 dB with the optimized trajectory, proving that mobility effectively eliminates the "edge user" disadvantage.
Conclusion
This work demonstrates the efficacy of 3D mobility in optimizing IoT data collection. By treating drone altitude and horizontal movement as dynamic variables, the system can adapt to user distributions and backhaul requirements. Future research will focus on integrating collision avoidance constraints and joint resource allocation for communication and computation in multi-drone networks.
Summary updated