Deep Learning Enabled Multicast Beamforming With Movable Antenna Array
Jae‐Mo Kang
Abstract
Beamforming with movable antenna (MA) array has recently attracted increasing attention as an enabling technology for next-generation wireless communications, e.g., 6G. In this paper, we consider a multicast scenario where a base station (BS) equipped with a linear MA array broadcasts common information to multiple users, each equipped with a single fixed-position antenna. Our objective is to jointly optimize the antenna position vector (APV) and antenna weight vector (AWV) by maximizing the minimum beamforming gain for the users, which is challenging to tackle analytically due to nonconvexity. To effectively and intelligently break through such challenge, we propose a novel deep learning (DL) model with three modules, namely, feature extractor, APV optimizer, and AWV optimizer. An effective training strategy for the proposed DL model is also developed in an unsupervised manner with a customized loss function. The superiority and effectiveness of the proposed scheme are confirmed through simulation results.