iRIS: A Deep Policy Learning Framework for Intelligent Distribution of RIS Elements

Sudhanshu Arya

doi:10.1109/OJCOMS.2025.3645826

Return

iRIS: A Deep Policy Learning Framework for Intelligent Distribution of RIS Elements

2025-01-01

PRE

Sudhanshu Arya

Saranya Karattupalayam Chidambaram

Yogesh Kumar Choukiker

Abhijit Bhowmick

DOI：10.1109/OJCOMS.2025.3645826

Original

Original request for help

Save

Abstract

En 中文

In the recent past, extensive research has been conducted considering reconfigurable intelligent surface (RIS) as an important building block in wireless communication systems. However, optimizing the RIS elements in multiuser scenarios is still an open challenge. Practically, under multiuser scenarios, partitioning of RIS elements among users is not an optimal and feasible solution, as any subgroup of elements will reflect everything it receives, not just the intended user’s signal. In this work, Ingenious Reconfigurable Intelligent Surface (iRIS), a novel and first-of-its-kind policy-driven framework designed to optimize the allocation of Reconfigurable Intelligent Surface (RIS) elements in wireless communication systems, is proposed. Leveraging an environment that models dynamic channel fading and user locations, the iRIS framework employs a deep neural network with a custom reward function as a policy network to predict optimal allocation strategies, dynamically adapting to varying channel conditions. With an adaptive learning mechanism equipped with neural policy approximation, iRIS enables intelligent allocation of RIS elements among multiple user pairs to maximize system performance for RIS-assisted dynamic channels. iRIS supports complex allocation strategies and allows for realistic modeling of RIS-user interactions. Additionally, a distribution function, coined SAYS, is derived to model the allocation of RIS elements, and provide insights into the statistical properties of the allocation process. The proposed machine learning-based RIS allocation provides low overhead and efficient multi-user communication. The results show that moderate phase quantization ensures practical deployability in real-world systems by striking a balance between complexity and performance. The sum-rate achieved demonstrates that the iRIS consistently maximizes communication performance while maintaining fairness across users in dynamic channels. Importantly, since the proposed framework involves only passive RIS elements, it can be effectively utilized in power-constrained systems.

Keywords:

RIS

policy-driven network

sum-rate