Research On Capacity Optimization Of MIMO-OFDM System Based On Intelligent Reflecting Surface

In recent years,the 5th generation mobile communication technology(5G)has developed vigorously.The Intelligent Reflecting Surfaces(IRS),composed of a large number of low-cost passive reflective elements,is one of the key technologies for 5G to improve system capacity.The reconfigurability of the IRS increases the design freedom of the communication system,while the passive nature and low-cost components of the IRS significantly reduce the cost of the communication system.This thesis mainly considers the performance optimization problem of IRSassisted multiple-input multiple-output(MIMO)orthogonal frequency division multiplexing(OFDM)millimeter wave(mmWave)downlink communication system.First,due to the energy loss of the reflection unit circuit,the IRS cannot fully reflect the incident signal.In this thesis,the actual IRS reflection coefficient phase shift model is considered.A novel joint design problem of the transmit beamforming and the IRS reflection coefficient matrix is proposed,constrained by the actual phase shift model,to maximize communication capacity.The optimization problem is decomposed into multiple sub-problems after equivalent transformation,and a joint design algorithm based on block coordinate descent(BCD)is proposed in this thesis.The simulation results demonstrate the effectiveness of the proposed algorithm.After introducing the actual phaseshift model,the system performance inevitably declines,but there is still a significant performance improvement compared to the case where IRS is not deployed.The considered optimization problem is closer to the real situation and further demonstrates the gain of IRS to the communication system.Secondly,in view of high complexity of traditional mathematical optimization methods,this thesis utilizes the idea of deep reinforcement learning,and transforms the original optimization problem into the form of deep reinforcement learning by reasonably mapping elements such as environment,state,action and reward,to explore the application of deep reinforcement learning in IRS-assisted communication optimization.Specifically,this thesis proposes a joint design framework based on the Soft Actor Critic(SAC)algorithm for the transmit beamforming and IRS reflection coefficients.The simulation results show that the performance of the proposed algorithm is close to the upper bound of theoretical performance and has extremely low computational complexity,which is of great significance for the deployment of the joint design algorithm in practical terminals.