Design and Optimization of Multi-Cell Multi-IRS Massive MIMO NOMA Networks

Abstract

We consider the downlink of a multi-cell massive multi-input-multi-output wireless network with an intelligent reflecting surface (IRS) in each cell. Each base station (BS) serves its users via its IRS by using non-orthogonal multiple access (NOMA). We derive a closed form spectral efficiency (SE) lower bound for our system by considering spatially-correlated Rician channels. This lower bound is then used to maximize the non-concave global energy efficiency (GEE) metric in two steps. In the first step, we optimize the BS power by leveraging the minorization-maximization (MM) framework, which transforms a non-concave optimization into a sequence of surrogate concave problems. We construct a novel surrogate function to develop the MM framework. In the second step, we design a novel low-complexity accelerated gradient projection algorithm to jointly optimize phases of IRS in all the cells. The algorithm accelerates its convergence by using Nesterov extrapolation. We analytically prove the algorithm convergence. We also show that for uncorrelated Rayleigh channels, SE becomes independent of the IRS phase. It is, therefore, crucial to consider spatial correlation while analyzing and developing IRS-based NOMA systems. We numerically investigate multiple aspects which are crucial for realizing tangible SE and GEE gains while designing multi-cell multi-IRS NOMA systems.