Soft actor critic-based performance optimization for IRS-aided cognitive radio systems

An intelligent reflective surface (IRS)-assisted cognitive radio (CR) multiple-input multiple-output (MIMO) communication system is considered. Incorporating cognitive radio and IRS capabilities into such a system yields significant improvements in system performance, including energy efficiency (EE) and receiver quality of service (QoS). For enhancing the attainable rate of secondary users (SU) without exceeding the interference temperature limit (IT) on the primary users (PU), a non-convex optimization problem is formulated, which is usually solved by means of alternative optimization (AO) methods such as block coordinate descent (BCD) algorithms. In this paper, we focus on deep reinforcement learning (DRL) approaches, specifically, the soft actor-critic (SAC) algorithm, to solve this optimization problem. For comparison, all simulation figures will be composed of a BCD benchmark beside the SAC curves. In addition, a 16-element MIMO antenna array for the secondary transmitter (ST) base station is proposed, designed, fabricated, and tested, yielding a 90% radiation efficiency with perfect impedance matching and acceptable return losses.