Deep Learning-Based Autonomous UAV-BSs for NGWNs: Overview and a Novel Architecture

To address the ever-growing connectivity demand in communications, the adoption of ingenious solutions, such as utilization of unmanned aerial vehicles (UAVs) as mobile base stations, is imperative. In general, the location of a UAV base station (UAV-BS) is determined by optimization algorithms, which have high computationally complexities and are hard to run on UAVs due to energy consumption and time constraints. In this article, we overview the UAV-BS trajectory optimization problem for next generation wireless networks and show that a convolutional neural network (CNN) model can be trained to infer the location of a UAV-BS in real time. To this end, we create a framework to determine the UAV-BS locations considering the deployment of mobile users (MUs) to generate labels by using the data obtained from an optimization algorithm. Performance evaluations reveal that once the CNN model is trained with the given labels and locations of MUs, the proposed approach is, indeed, capable of approximating the results given by the adopted optimization algorithm with high fidelity, outperforming reinforcement learning-based approaches in resource-constrained settings. We also explore future research challenges and highlight key issues.

Automated summary

To address the growing demand for connectivity in communications, it is necessary to adopt innovative solutions such as using unmanned aerial vehicles (UAVs) as mobile base stations. This article presents an overview of the UAV base station trajectory optimization problem for next generation wireless networks and demonstrates that a convolutional neural network (CNN) model can be trained to accurately infer the location of a UAV base station in real time. Performance evaluations show that the proposed approach, trained with given labels and mobile user locations, can approximate the results of the optimization algorithm with high fidelity and outperform reinforcement learning-based approaches in resource-constrained settings. Future research challenges and key issues are also discussed.