Data-driven online learning algorithm for optimal linear tracking control over unreliable wireless MIMO fading channels

This work explores the data-driven online tracking control problem for linear dynamic systems across multiple-input multiple-output (MIMO) fading channels. Initially, we address the optimal tracking control for a system with known plant dynamics, and design an innovative stochastic-approximation (SA)-based data-driven algorithm that leverage the instantaneous wireless channel state information (CSI). Subsequently, we extend this approach to accommodate unknown plant dynamics by proposing a novel normalized-stochastic-gradient-descent (NSGD)-based algorithm. This algorithm facilitates simultaneous system identification and control in an online setting using the real-time plant state as well as the CSI. Through Lyapunov drift analysis, we establish the asymptotic optimality of our proposed data-driven algorithms. Numerical results and analysis further demonstrate notable performance improvements compared to several leading learning techniques.