Deep-Reinforcement-Learning-for-Intelligent-reflecting-Surfaces

Intelligent Reflecting Surfaces (IRSs): A Promising Technology for 6G Networks: In this work, we examine a downlink MISO scenario with an intelligent reflecting surface (IRS) to maximise the SNR for the user. The IRS optimization problem is complex and non-convex because it necessitates the tuning of the phase shift reflection matrix with unit modulus constraints. We use deep reinforcement learning (DRL) to forecast and optimally adjust the IRS phase shift matrices, owing to the increasing use of DRL approaches capable of tackling non-convex optimization problems. The Deep deterministic policy gradient (DDPG) algorithm in [1] has been studied and implemented in Python from the scratch. Te IRS-assisted MISO system based on the DRL scheme produces a high SNR, according to implementation and simulation results. Furthermore, we modify the IRS-DRL framework to account for unknown channel gains between IRS and user, which would occur in practice because IRS is uninformed of the user’s position and cannot forecast the angles of departures, as required by the assumed Rician Channel fading model.