Deep Unfolded Multi-Group Multicast Beamforming.

The multicast beamforming problem with qualityof-service constraints is an NP-hard quadratically constrained quadratic program, for which a variety of low-complexity approximation techniques have been proposed. A recent approach to multicast beamforming relies on an algorithm with projections onto convex sets and bounded perturbations to approximate solutions in a Hilbert space. Owing to its iterative nature, this algorithm is suited for deep unfolding, where its design parameters are learned using some training data. A deep unfolded multicast beamforming algorithm for the single-group setting has recently been proposed. However, this method is not applicable in multi-group settings. In this paper, we propose a deep-unfolded algorithm for multicast beamforming in a multi-group setting, which is based on an algorithm with superiorized projections onto convex sets. Simulations show that the proposed method outperforms the underlying untrained algorithm in the multi-group setting. At the same time, its performance matches that of the existing deep-unfolded single-group multicast beamforming algorithm if there is only one multicast group.