Multimodal sensing and DRL-driven beam selection in RIS-aided mmWave mMIMO systems

Kanaan, Khalid, Nasser, Ahmed, Celik, Abdulkadir, Wang, Ruiqi, Shamim, Atif and Eltawil, Ahmed M. (2025) Multimodal sensing and DRL-driven beam selection in RIS-aided mmWave mMIMO systems. In 2025 IEEE 36th International Symposium on Personal, Indoor and Mobile Radio Communications, PIMRC 2025. IEEE. 6 pp . (doi:10.1109/PIMRC62392.2025.11274583).

Abstract

The IMT-2030 vision emphasizes two key 6G directions: integrated sensing and communication (ISAC) alongside artificial intelligence (AI)-native frameworks, where multimodal sensory data inputs enhance situational awareness and adaptive decision-making of communication systems. Accordingly, this paper introduces a deep reinforcement learning (DRL)-based beam selection framework for downlink multi-user reconfigurable intelligent surface (RIS)-assisted millimeter-wave (mmWave) massive MIMO (mMIMO) systems. Targeting maximized sum rates under quality of service (QoS) and fairness constraints, the framework employs two primary sensing modalities: a stereo camera mounted on the RIS for user equipment (UE) detection and inertial measurement units (IMUs) on UEs to obtain 3D Cartesian coordinates, thus eliminating the need for channel state information (CSI) acquisition. The DRL framework combines two algorithms - double deep Q-network (DDQN) and proximal policy optimization (PPO) - to jointly optimize RIS phase shifts and UE receive beamformers through predefined codebooks and adaptive beam selection. A testbed was developed to validate the system, leveraging real-world data to train the DRL algorithms. Experimental results demonstrate that both agents achieve nearoptimal sum rates across diverse base station (BS) transmit power levels and QoS thresholds while reducing computational complexity by 95%, illustrating the framework's potential for efficient and scalable beam alignment for AI-native wireless systems.

This record has no associated files available for download.

Contributors

Author: Abdulkadir Celik

Download statistics