Research On Joint Beamforming Algorithm Of Distributed IRS Assisted Millimeter-Wave MU-MISO Communication System

The 6th generation(6G)wireless communication is one of the key enablers of the future smart digital society,which can provide quality communication services such as ultra-high data rates,high energy efficiency,wide global coverage,and high security.To meet these service needs,various technologies have been proposed in recent years,such as millimeter wave(mm Wave)communications,massive multiple input multiple output(m MIMO)antenna arrays,and intelligent reflecting surfaces(IRS).IRS enables an intelligent and reconfigurable propagation environment in wireless communication systems with low cost and low energy consumption,and by deploying IRS in wireless communication,the radio channel between transmitter and receiver is reconfigured to solve the problem of m MIMO system beamforming.Communication interruptions in mm Wave non-line-of-sight scenarios.Therefore,the joint beamforming design problem for the downlink of a distributed IRS-assisted millimeter wave multi-user-multiple input single output(MU-MISO)system is investigated under the premise that mm Wave,beamforming,and IRS are complementary to each other.The main contributions of the paper are divided into three areas as follows.1.A distributed intelligent reflecting surface-assisted joint beamforming design algorithm under the millimeter-wave MU-MISO system(BCD-P)is proposed for the reliable communication problem of the downlink of the millimeter-wave MU-MISO system in the non-line-of-sight scenario.Assuming perfect channel state information and the use of salehvalenzuela(S-V)millimeter-wave channel model,the algorithm models the joint solution of active beamforming(ABF)at base stations(BS)and passive beamforming(PBF)at multiple smart reflecting surfaces as a nonconvex optimization problem with the objective of maximizing the multi-user weighted sum rate(WSR)under power and constant mode constraints.The joint optimization problem is decoupled into an easily solvable equivalence problem using a closed-form fractional programming technique.The convergence of the BCD-P algorithm is proved and the time complexity analysis is given.The convergence proof and time complexity analysis of the BCD-P algorithm are also given.Simulation experiments are conducted to analyze the impact of IRS hardware limitations and other factors on the performance of the algorithm.The simulation results show that the BCD-P algorithm can converge quickly and can effectively increase the system transmission rate with reduced complexity compared to the two baseline algorithms.2.To address the problems of channel estimation error due to the passive nature of IRS and the performance degradation and non-convergence of the BCD-P algorithm in the case of imperfect channel state information,and improved BCD alternate optimization algorithm(BCD-I)is proposed.Assuming the case of imperfect channel state information and model the S-V channel model with estimation error,the algorithm also considers the power and constant mode constraints and aims at the maximum multi-user WSR.The closed-form fractional programming technique is used to decouple and optimize the ABF and PBF matrices alternately according to the approximate linear rule and the improved distributed stochastic continuous convex approximation rule,respectively.And the time complexity analysis of the BCD-I algorithm is given.The simulation results show that the BCD-I algorithm can converge in the case of imperfect channel state information and significantly improve the system transmission rate compared with the BCD-P algorithm.3.To address the problems of the high computational complexity of traditional algorithms and the need for continuous iterative solutions during operation,a generalized deep reinforcement learning algorithm for joint beamforming based on proximal policy optimization(PPO)is proposed to jointly solve the ABF and PBF matrices.The algorithm considers two cases of perfect and imperfect channel state information,defines power and constant mode constraints in a deep reinforcement learning environment,uses WSR as an environmental reward to train the PPO model,and the intelligent body generates continuous ABF and PBF matrices by the obtained action distributions.A joint transfer learning-proximal policy optimization(TL-PPO)generalized beamforming algorithm is proposed to address the problem of system performance degradation when the IRS position is changed.The algorithm uses a model-based migration learning algorithm to improve the training efficiency.And the time complexity analysis of the PPO and TL-PPO algorithms is given separately.Simulation results show that the PPO algorithm can converge and outperform the BCD-I and BCD-P algorithms,the degree of imperfection of the channel state information does not affect the algorithm performance,and the TL-PPO algorithm can reduce the training time overhead of the model by 35% when the environment changes.