Scholarlink

<p>This letter addresses the critical challenge of dynamic link blockages in 6G Terahertz (THz) communications, which severely degrade transmission reliability and spectral efficiency. We propose a meta-reinforcement learning (Meta-RL) framework for beam optimization in Reconfigurable Intelligent Surface (RIS)-assisted systems. By exploiting the geometric properties of the approximation loss, our method designs a novel regularization term that enables the beam controller to adapt rapidly to new channel environments with minimal gradient updates (as few as 18 steps). Simulation results demonstrate that the proposed approach achieves a 35% throughput gain over standard deep reinforcement learning algorithms and a 2.3-fold improvement over codebook-based schemes at 10 dB SNR, while reducing convergence time by an order of magnitude (from over 150 steps to under 20 steps). These quantifiable advancements offer a robust and efficient beam management solution for dynamic THz communication scenarios, addressing the millisecond-level adaptation requirements of 6G systems.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Meta-Reinforcement Learning for Fast Beam Adaptation in 6G Terahertz Reconfigurable Intelligent Surface Communications

  • 期刊论文

  • 发表日期：

    2026年2月4日

查看全文
  • Chaoran Du,
  • Letian Huang

摘要

This letter addresses the critical challenge of dynamic link blockages in 6G Terahertz (THz) communications, which severely degrade transmission reliability and spectral efficiency. We propose a meta-reinforcement learning (Meta-RL) framework for beam optimization in Reconfigurable Intelligent Surface (RIS)-assisted systems. By exploiting the geometric properties of the approximation loss, our method designs a novel regularization term that enables the beam controller to adapt rapidly to new channel environments with minimal gradient updates (as few as 18 steps). Simulation results demonstrate that the proposed approach achieves a 35% throughput gain over standard deep reinforcement learning algorithms and a 2.3-fold improvement over codebook-based schemes at 10 dB SNR, while reducing convergence time by an order of magnitude (from over 150 steps to under 20 steps). These quantifiable advancements offer a robust and efficient beam management solution for dynamic THz communication scenarios, addressing the millisecond-level adaptation requirements of 6G systems.