An ML-Assisted OFDM-Based Hemispherical Array Antenna with Hybrid Beamforming for HAPS

A high-altitude platform station (HAPS) located in the stratosphere can provide connectivity over a large area. However, a HAPS can create only a limited number of uncorrelated beams. Therefore, we cannot dedicate a beam to each user, and we need to perform user scheduling to be able to serve a large number of users with a HAPS. To this end, in this paper, we group users into clusters using the K-means algorithm and allocate the users in each cluster to orthogonal frequency resource blocks. The frequencies are reused across the clusters. Since HAPS has limited power resources, we also propose a novel codebook-based hybrid beamforming scheme. The results of our simulations conclusively show that our proposed beamforming scheme outperforms the traditional steering vector-based beamforming scheme in high-rise urban areas. Furthermore, we propose a deep Q-network (DQN)-based power allocation method that considerably outperforms the baseline equal power allocation scheme for large coverage areas. Finally, we compare the interference management efficiency of our proposed orthogonal frequency-division multiplexing (OFDM)-based hybrid beamforming-enabled hemispherical array antenna and baseline rectangular and cylindrical array antennas.