Virtual Full-Duplex Communication Enabled STAR-RIS: Performance Analysis and Machine Learning-Assisted Optimization

This work proposes a novel virtual full-duplex (VFD) communication based simultaneously transmitting and reflecting reconfigurable intelligent surfaces (STAR-RIS) scheme to mimic as well as outperform conventional full-duplex based STAR-RIS communication in practical high residual self-interference scenarios. For the proposed VFD based STAR-RIS (VFD-STAR-RIS) system involving multiple uplink and downlink users with user selection, analytical expressions of outage probability (OP) and ergodic rate (ER) are presented for downlink and uplink users. Thereafter, to minimize STAR-RIS aided inter-user interferences, we study joint optimization of user power allocations, reflection amplitude, transmission amplitude and element partitioning (JPRTE) for system OP (SOP) minimization and ergodic sum rate (ESR) maximization. A particle swarm optimization (PSO) based solution is used to solve the JPRTE problem of minimizing the SOP (JPRTE-SOP). However, due to the complexity involved in the ER expressions, applying PSO directly to the JPRTE problem of ESR maximization (JPRTE-ESR) will require significant convergence time. Thus, a machine learning (ML) based solution is proposed where the ER expressions are first closely approximated via a ML architecture, and thereafter PSO is applied to obtain a solution having a very low computational time. Monte-Carlo simulations are carried out to demonstrate efficacy of proposed VFD-STAR-RIS scheme, JPRTE-SOP, and JPRTE-ESR solutions to draw out useful inferences.