Scholarlink

<p>Semantic communication serves as an effective approach for real-time perception in next-generation wireless communication systems. In this paper, we construct a multi-UAV semantic monitoring system with adaptive semantic compression strategy. For video frame transmission in this system, we introduce a novel metric called the Age of Incorrect Semantics (AoIS). Then, we formulate an optimization problem to minimize the total AoIS by jointly optimizing the compression ratio, transmission power, and transmission decisions. Since semantic encoding and decoding cannot be expressed by explicit formulas, conventional convex optimization iterative algorithms cannot solve this problem. To solve this challenging problem, we develop a deep reinforcement learning (DRL) algorithm based on centralized training with decentralized execution (CTDE), where global observations are employed for both critic network training and advantage computation, while local observations are used for policy execution. Simulation results demonstrate that the proposed strategy effectively balances semantic similarity and transmission frequency compared with the baselines.</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

Joint optimization of adaptive semantic compression and resource allocation for semantic-enabled UAV communications

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月29日

查看全文
  • Yuchen Zhou,
  • Peng Dai,
  • Bingtao He,
  • Mengmeng Ren,
  • Mofan Luo,
  • Jian Chen

摘要

Semantic communication serves as an effective approach for real-time perception in next-generation wireless communication systems. In this paper, we construct a multi-UAV semantic monitoring system with adaptive semantic compression strategy. For video frame transmission in this system, we introduce a novel metric called the Age of Incorrect Semantics (AoIS). Then, we formulate an optimization problem to minimize the total AoIS by jointly optimizing the compression ratio, transmission power, and transmission decisions. Since semantic encoding and decoding cannot be expressed by explicit formulas, conventional convex optimization iterative algorithms cannot solve this problem. To solve this challenging problem, we develop a deep reinforcement learning (DRL) algorithm based on centralized training with decentralized execution (CTDE), where global observations are employed for both critic network training and advantage computation, while local observations are used for policy execution. Simulation results demonstrate that the proposed strategy effectively balances semantic similarity and transmission frequency compared with the baselines.