Reconfigurable intelligent surface beam tracking based on integrated control architecture and vision‑aided sensing for mobile wireless communication

Abstract

The traditional RIS beam tracking system primarily relies on signal feedback from the UE (User Equipment), transmitting feedback data via TCP and performing an exhaustive scan of a predefined codebook to determine the optimal beam direction. However, such methods involve interactions among multiple devices and extensive codebook traversal, leading to significant system latency. To address this issue, this study first designs an integrated RIS control architecture that consolidates the UE and RIS control terminal into a single server while assigning a dedicated IP address to the RIS via a router, enhancing data transmission stability and efficiency. Experimental results show that, compared to traditional methods, this approach reduces beam tracking latency by more than 50% and ensures stable signal reception and demodulation in non-line-of-sight (NLOS) environments. Building on this foundation, a vision-based RIS beam control scheme is further proposed. This scheme utilizes the YOLOv7 deep learning model in combination with an Intel RealSense D435i depth camera for user detection, directly identifying the target region and selecting the corresponding RIS codebook for beamforming. This eliminates the feedback and exhaustive scanning overhead present in traditional methods. Experimental results indicate that the vision-driven RIS approach completely removes feedback and scanning delays, achieving a latency of approximately 0.3s. Even in wireless propagation NLOS scenarios the system maintains stable signal reception and demodulation.