Research On Beamforming And Channel Tracking For 6G-Oriented RIS-Assisted Communication Systems

A ubiquitous,high-throughput,green,and safely controllable intelligent mobile communication system is an important guarantee to support the digital development of economic society.Reconfigurable intelligent surface(RIS),being envisioned as an emerging wireless communication technology is expected to catalyze a new paradigm shift to the sixth generation mobile communication system(6G)for easy deployment,low power consumption,and programmable,which is qualified for the growing communication requirements.However,there still exist numerous challenges regarding the co-design of the active and passive beamforming,as well as the acquisition of the cascaded channel state information(CSI)in RIS-assisted wireless communication systems.Explicitly,on the one hand,the doppler-induced time varying channels significantly degrades the performance of beamforming schemes in highmobility scenarios,such as the Internet of Vehicles(IoV)and high-speed train communications.Frequent acquisition of the full channel state information may entail excess pilot overhead,sacrificing the transmission time.To tackle this issue,a robust beamforming scheme is conceived with reduced pilot overhead and complexity.On the other hand,the near-field radiation effect cannot be ignored when the scale of the RIS increases,in which the channel matrix no longer exhibits sparsities in the angular domain.Hence,it is necessary to explore the sparsity of the extremely large-scale RIS(XL-RIS)cascaded channel,and a sparsity reconstruction algorithm is necessary.To address the aforementioned challenges,firstly,this thesis outlines the relevant theoretical foundations of the work,including the time-varying channel model,sparse angular domain transformation of the multiple input multiple output(MIMO)millimeter-wave channel and RIS cascaded channel,the basic working principle of RIS,and compressive sensing technology.Next,the thesis conducts research on the beamforming design for a RIS-assisted high-mobility system in the presence of the channel aging effect.Particularly,a novel frame structure is devised,which divides the estimation of the slow-changing angles of arrival/departure(AoAs/AoDs)and fast-changing small-scale fading characteristics into two stages in an attempt to alleviate the excess pilot overhead.Considering the channel correlation between two adjacent instants,the first-order autoregressive model is applied to characterize the statistical characteristics of the smallscale fading,and then a long-term optimization objective is established.Furthermore,a low-complexity joint beamforming optimization algorithm is proposed based on the deep deterministic policy gradient(DDPG)network.The proposed algorithm can adjust the active and passive beamforming strategy dynamically based on the partial CSI determined by AoAs/AoDs,that can be readily acquired in practice,initial CSI,and realtime environment feedback,with the goal of mitigating the impact of the channel aging effect.Simulation results show that the proposed algorithm reduces the performance gap between the benchmark and the ideal upper bound by 27%,and verify the robustness of the proposed algorithm.Then,the thesis investigates the near-field cascaded channel estimation for an XL-RIS-assisted wireless communication system.Based on the sparse XL-MIMO channel in the polar domain,this thesis derives the sparse-domain representation of the XL-RIS cascaded channel.A twostep sparse reconstruction algorithm is proposed based on deep learning networks.In the first step,a denoising convolutional neural network(DnCNN)is used to estimate the nonzero rows of the sparse channel matrix,where the nonzero rows correspond to the common arrival parameters of the base station(BS)in the cascaded channel.Due to the presence of convolutional layers,the power leakage is effectively mitigated.In the second step,the neural network based on the Iterative Shrinkage Thresholding Algorithm(ISTA)is proposed to adaptively learn the dictionary matrix and channel sparse domain representation,which avoids the power drift induced by the column coherence of the dictionary matrix.Simulation results show that high estimation accuracy can be achieved with sufficient pilots and high signal-to-noise ratio.