Synaptic Intelligence-Based Beam Selection in Dynamic Environments
Yunwei Gou, Yawen Chen, Yifan Zhu, Wan Xiang, Zhaoming Lu, Xiangming Wen · IEEE Communications Letters · 2025
In real-world vehicular communications, machine learning-based beam selection is challenging under non-stationary distributions of Non-Line of Sight (NLOS) and Line of Sight (LOS) cases. For example, a model newly updated on rush-hour traffic hardly re-adapts to previously encountered regular traffic. To address this, the letter proposes a continual learning approach, named Synaptic Intelligence-based Beam Selection (SIBS), which retains historical knowledge by restricting the change to key parameters during the new training. The experiments on simulated datasets show strong adaptability of SIBS to dynamic environments, where it adapts to the new scenario and notably maintains performance over the encountered scenario.