Using STAR-IRS to Secure Indoor Communications Through Symbol-Level Random Phase Modulation
Du, Sun, Zhang et al.
This paper proposes a secure indoor communication scheme based on simultaneous transmitting and reflecting intelligent reflecting surface (STAR-IRS). Specifically, a transmitter (Alice) sends confidential information to its intended user (Bob) indoors, while several eavesdroppers (Eves) lurk outside. To safeguard the transmission from eavesdropping, the STAR-IRS is deployed on walls or windows. Upon impinging on the STAR-IRS, the incoming electromagnetic wave is dynamically partitioned into two components, enabling both transmission through and reflection from the surface. The reflected signal is controlled to enhance reception at Bob, while the transmitted signal is modulated with symbol-level random phase shifts to degrade the signal quality at Eves. Based on such a setting, the secrecy rate maximization problem is formulated. To solve it, a graph neural network (GNN)-based scheme is developed. Furthermore, a field-programmable gate array (FPGA)-based GNN accelerator is designed to reduce computational latency. Simulation results demonstrate that the proposed strategy outperforms both the conventional scheme and the reflection-only scheme in terms of secrecy performance. Moreover, the GNN-based approach achieves superior results compared to benchmark techniques such as maximum ratio transmission (MRT), zero forcing (ZF), and minimum mean square error (MMSE) in solving the optimization problem. Finally, experimental evaluations confirm that the FPGA-based accelerator enables low inference latency.