Cell-Free Massive MIMO SWIPT with Beyond Diagonal Reconfigurable Intelligent Surfaces

TL;DR

This paper explores integrating beyond diagonal reconfigurable intelligent surfaces into cell-free massive MIMO systems to significantly improve simultaneous wireless information and power transfer, using advanced optimization and machine learning techniques.

Contribution

It introduces a novel BDRIS-assisted CFmMIMO framework with joint optimization of AP selection, power control, and BDRIS design, achieving substantial energy harvesting improvements.

Findings

Up to sevenfold increase in harvested energy with BDRIS.

Efficient algorithms for joint optimization are proposed.

Significant spectral efficiency and energy harvesting gains achieved.

Abstract

We investigate the integration of beyond diagonal reconfigurable intelligent surfaces (BDRISs) into cell free massive multiple input multiple output (CFmMIMO) systems to enhance simultaneous wireless information and power transfer (SWIPT). To simultaneously support two groups of users energy receivers (ERs) and information receivers (IRs) without sacrificing time frequency resources, a subset of access points (APs) is dedicated to serving ERs with the aid of a BDRIS, while the remaining APs focus on supporting IRs. A protective partial zero forcing precoding technique is implemented at the APs to manage the non coherent interference between the ERs and IRs. Subsequently, closed form expressions for the spectral efficiency of the IRs and the average sum of harvested energy at the ERs are leveraged to formulate a comprehensive optimization problem. This problem jointly optimizes the AP selection, AP power control, and scattering matrix design at the BDRIS, all based on long term statistical channel state information. This challenging problem is then effectively transformed into more tractable forms. To solve these sub problems, efficient algorithms are proposed, including a heuristic search for the scattering matrix design, as well as successive convex approximation and deep reinforcement learning methods for the joint AP mode selection and power control design. Numerical results show that a BDRIS with a group or fully connected architecture achieves significant energy harvesting gains over the conventional diagonal RIS, especially delivering up to a seven fold increase in the average sum of harvested energy when a heuristic based scattering matrix design is employed.