Deep Reinforcement Learning Assisted Spectrum Management in Cellular Based Urban Air Mobility

The emerging urban air mobility (UAM) opens a new transportation paradigm to support increasing mobility demand in metropolitan areas. A major challenge for UAM is to ensure reliable two-way wireless communications between aerial vehicles and their associated ground air traffic control centers for safe operations. The concept of cellular-based UAM (cUAM) provides a promising solution for reliable air-ground communications in urban air transportation, where each aerial vehicle is integrated into an existing cellular network as a new aerial user, sharing the cellular spectrum with existing terrestrial users. Generally, the additional aeronautical use of cellular spectrum can introduce harmful interference to current terrestrial communications, which only amplifies the severity of spectrum scarcity issues. Therefore, a new spectrum management solution is necessary for cUAM applications. In this article, we first introduce the communication requirements and spectrum management challenges in cUAM. Then we propose to apply deep reinforcement learning technology to perform dynamic spectrum management in cUAM. Next, a cUAM use case is investigated where a deep-reinforcement-learning-based dynamic spectrum sharing solution is proposed to minimize the total UAM mission completion time. Numerical results show that the proposed solution can reduce the mission completion time and improve the spectrum utilization efficiency. Finally, we present several directions for future research.