A User Fairness-Aware Optimization of IRS-Assisted MIMO Communications

In this work, we approach optimizing intelligent reflecting surface (IRS)-assisted multiple-input multiple-output (MIMO) communications as a weighted multi-objective optimization problem with two aims: maximizing the system’s sum-rate and ensuring fairness in user rates. We employ a conventional weighted sum-rate function to pursue the first objective, while the second one is achieved by minimizing the maximum weighted mean squared error (MSE) value. We formulate this task as a contextual bandit (CB) problem and leverage the deep contextual bandit-oriented deep deterministic policy gradient (DCB-DDPG) framework to jointly determine the IRS phase shifts and precoder matrices that enhance the system performance across various configurations.