Meta-Learning-Based STAR-RIS for Dynamic Multi-Mobile-User Downlink Communications Over 6G Mobile Wireless Networks

Reconfigurable intelligent surface(RIS) has been widely envisioned as a key technique, which can enhance communicationquality of service(QoS) formobile users(MUs) by reconfiguring wireless propagation environments. Unlike traditional RISs that can only reflect signals,simultaneously transmitting and reflecting RIS(STAR-RIS) can extend half-space coverage to full-space coverage by simultaneously transmitting and reflecting incident signals. In this paper, we propose to develop the joint optimization for the transmission and reflection coefficients, i.e., phase-shifts and amplitudes, of STAR-RIS and the transmit beamforming ofbase station(BS) over STAR-RIS-aided downlink communications, where multiple MUs move in real time and can change their movement patterns dynamically. First, taking into account MUs’ mobility, we formulate a rate maximization problem to maximize the average transmission rate from BS to MUs, under the continuous phase-shifts and path-loss channel modes with perfectchannel state information(CSI)/MUs-locations estimation. Second, since themeta-learningcan rapidly adapt to dynamic network environments, we develop a scheme based on meta-learning integrated with theproximal policy optimization(PPO) algorithm to optimize STAR-RIS’s transmission and reflection coefficients and BS’s transmit beamforming. Third, we extend our work to the more realistic communication scenario under the discrete phase-shifts and Rician channel models with imperfect CSI/MUs-locations estimation. Finally, we validate and evaluate our developed schemes through extensive simulations, showing that our meta-learning-based schemes significantly outperform baseline schemes and can adapt to MUs’ new movement patterns rapidly.