Research On Reflection Pattern Configuration And Receiver Design In Reconfigurable Intelligent Surface-assisted MIMO System

Reconfigurable Intelligent Surface(RIS)is considered as a key technology of future wireless communication,enabling free regulation of the radio propagation environments.However,the fact that the RIS is passive device,which only reflect their impinging signals in a configurable manner,gives rise to a multitude of signal processing challenges,including the method of symbol detection。Because of the high cost of obtaining Channel State Information(CSI),a large number of datadriven receiver applications have been stimulated,aiming to simulate the channel input-output relationship through the training of a large amount of data to achieve reliable communication.Datadriven receivers are trained to acquire implicit information from data and avoid direct acquisition and utilization of CSI.This thesis studies the optimization of RIS and receiver learning in Multiple Input Multiple Output(MIMO)system assisted by the RIS,and the main contents are as follows:(1)For the uplink of the RIS-assisted MIMO system,a receiver based on the Deep Neural Network(DNN)is deployed at the base station to realize decoding.In this scenario,the application of Bayesian Optimization(BO)is motivated by treating the configuration of RIS reflection patterns that affect the underlying propagation channel as hyperparameters.Based on the background,a twostage optimization method is proposed to jointly optimize RIS and DNN-based receiver,using Bayesian Optimization and gradient descent,respectively.And a bayesian machine learning framework for optimizing RIS and network parameters is proposed,according to which the transmit pilots are directly used to jointly tune RIS and multi-antenna receiver.For the parameter optimization with limited dimension,the RIS grouping method is used to extend the low dimension to the high dimension.The simulation results show the effectiveness of the proposed framework for system performance improvement.(2)In-depth study of the learning method for the DNN-based receivers,aiming to reduce the training cost and improve the performance for the receivers.Solve problems such as periodic transmission of pilots that cause the network to be trained frequently,require a large amount of training data,training time,and slow network convergence.According to the learning idea of metalearning,the system model is constructed task-based.A receiver learning algorithm including metatraining,meta-adaptation and testing is proposed.In the process of training,the learning process is optimized,including gradient approximation and parameter inheritance,which improves the efficiency of inner update and outer update in the meta-training stage,and the convergence speed and generalization ability of the network in the adaptation stage.For the experimental design,the simulation results of the traditional receiver under different channel models show that the proposed learning algorithm for the DNN-based receiver improves the convergence,training sample size and system performance.