Near-Field Deployment Optimization of Hybrid Active-Passive RISs for Multi-BS Large-Scale Networks

Reconfigurable intelligent surface (RIS) has emerged as a promising technology for boosting system performance by intelligently controlling the radio environment. The utilization of large-scale RIS arrays and high-frequency signaling inevitably make near-field effects dominant in wireless communications. Hybrid active-passive RIS, with both active and passive reflectors, can reflect and amplify incident signals, significantly improving transmission performance while lowering overall power demands. This paper delves into a large-scale network scenario wherein numerous hybrid active-passive RISs lend aid to enhancing near-field multi-base station (BS) communication in the presence of signal obstructions. Our primary objective centers on the pursuit of maximizing the network ergodic capacity through a joint optimization of RIS deployment locations, panel orientation angles, and reflecting coefficients, thereby ensuring a stable network layout. To do this, the ergodic capacity is initially derived to achieve a tractable form by leveraging Jensen's inequality and binomial expansion theorem. Then, we address the challenges of the high-dimensional optimization problem within the soft actor-critic (SAC) network framework rooted in deep reinforcement learning (DRL). Remarkably, this proposal achieves near-optimal performance with low complexity, revealing that the best placement for RISs to maximize ergodic capacity is not necessarily close to the BS or users. Instead, it directly correlates with regions where users encounter signal obstruction, and where RISs must also be precisely oriented to establish robust virtual links for affected users. Simulation results demonstrate prominent performance enhancement over other benchmark deployment strategies, thereby providing an efficient deployment guideline of RIS arrays for multi-BS large-scale networks.