Dual-Polarized Multi-Cell Massive MIMO IRS Systems: SE Analysis, Low-Complexity Power, and Deep-Phase Optimization

We consider a downlink multi-cell massive multi-input-multi-output system, where the base station (BS) in each cell serves its user equipments (UEs) via a cell-specific intelligent reflecting surface (IRS). The BS, IRS and UEs have dual-polarized antennas. We derive a lower bound on the spectral efficiency (SE) of this system, which operates in practical spatially-correlated Rician-fading channels. We also maximize the system global energy efficiency by optimizing the BS transmit powers and IRS phases. We optimize the transmit powers by designing a low-complexity minorization-maximization algorithm, which provides a closed form solution. We then optimize the IRS phases by using a multi-agent deep deterministic policy gradient method, where each BS learns optimal phases by sharing only the policies with each other, and that too without any supervision. We numerically show i) the BS-UE link attenuation for which an IRS can be replaced with dual-polarized antennas at the BS and UEs; and ii) that dual-polarized antennas enable a multi-cell IRS system to match its SE with a single-cell system with single-polarized antennas at the BS, IRS and UEs.