Generative AI-Driven Phase Control for RIS-Aided Cell-Free Massive MIMO Systems

This work investigates a generative artificial intelligence (GenAI) model to optimize the reconfigurable intelligent surface (RIS) phase shifts in RIS-aided cell-free massive multiple-input multiple-output (mMIMO) systems under practical constraints, including imperfect channel state information (CSI) and spatial correlation. We propose two GenAI based approaches, generative conditional diffusion model (GCDM) and generative conditional diffusion implicit model (GCDIM), leveraging the diffusion model conditioned on dynamic CSI to maximize the sum spectral efficiency (SE) of the system. To benchmark performance, we compare the proposed GenAI based approaches against an expert algorithm, traditionally known for achieving near-optimal solutions at the cost of computational efficiency. The simulation results demonstrate that GCDM matches the sum SE achieved by the expert algorithm while significantly reducing the computational overhead. Furthermore, GCDIM achieves a comparable sum SE with an additional $98%$ reduction in computation time, underscoring its potential for efficient phase optimization in RIS-aided cell-free mMIMO systems.

💡 Research Summary

This paper addresses the challenging problem of phase‑shift optimization for reconfigurable intelligent surfaces (RIS) in cell‑free massive MIMO systems, where practical constraints such as imperfect channel state information (CSI) and spatial correlation are taken into account. Traditional optimization techniques—exhaustive search, block coordinate descent, successive convex approximation, or meta‑heuristics like genetic algorithms—either become computationally prohibitive or yield sub‑optimal solutions when the number of RIS elements grows large. To overcome these limitations, the authors propose two generative artificial intelligence (GenAI) approaches based on diffusion models: a Generative Conditional Diffusion Model (GCDM) and a Generative Conditional Diffusion Implicit Model (GCDIM).

The system model consists of M distributed access points (APs), K single‑antenna users, and an RIS with N passive reflecting elements. Channels from APs to RIS and from RIS to users are modeled with large‑scale fading coefficients (β) and spatial correlation matrices, while direct AP‑user links are also considered. Uplink training uses orthogonal pilots and linear minimum‑mean‑square‑error (LMMSE) estimation; downlink transmission employs conjugate beamforming based on statistical CSI. A closed‑form lower bound on the achievable sum spectral efficiency (SE) is derived using the use‑and‑then‑forget technique, leading to a highly non‑convex optimization problem over the phase vector θ ∈