FPGA Based Adaptive Beamforming for RIS using Deep Learning

Abstract

Reconfigurable intelligent surface (RIS) has emerged as a transformative technology for the next generation wireless communication systems, enabling dynamic control over propagation environment through programmable metasurface. However, implementing an adaptive beamforming technique for RIS hardware platform, such as Field Programmable Gate Arrays (FPGAs), presents significant challenges due to high computational complexity and resource demands of traditional techniques, including minimum variance distortionless ratio (MVDR). This paper proposes a novel convolutional neural network (CNN) assisted beamforming framework tailoring for smart RIS, especially optimized for FPGA deployment. By replacing matrix intensive computations with learned representations, the CNN model infers optimal phase shift configurations directly from input features such as spatial signal distributions or sensor measurements. The proposed method is compared with the conventional beamforming approaches, and the results demonstrate that the CNN-based beamforming significantly reduces the interference time and logic resource consumption while maintaining adaptability, stability, and accuracy, highlighting its potential for real time, low power RIS control. This work lays the foundation for intelligent and hardware efficient beamforming strategies suitable for advanced programmable wireless environments. © 2025 IEEE.