Scholarlink

<p>To address the challenges of slow convergence in large state-action spaces, limited utilization of prior knowledge, and constrained computational resources at terminal communication devices faced by reinforcement learning-based anti-jamming algorithms, this paper proposes a novel knowledge-assisted reinforcement learning algorithm for anti-nonlinear periodic jamming. The proposed algorithm enables communication receivers to sense channel jamming information to estimate whether dynamic jamming exhibits periodicity and to identify its period and pattern, thereby extracting anti-jamming knowledge. This knowledge is then embedded into the Q-table for initialization, with multi-value learning and pruning operations subsequently employed to further enhance the learning efficiency. Simulation results demonstrate that, under periodic dynamic jamming environments, the proposed knowledge-assisted multi-value reinforcement learning algorithm enables rapid response to evade jamming signals. Compared with conventional reinforcement learning methods without knowledge assistance, the proposed algorithm reduces the time required for effective anti-jamming decision-making by 90.9% to 93.9%, while achieving throughput improvements of 71.4% to 77.8% in dynamic jamming environments.</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

A novel knowledge-assisted reinforcement learning algorithm for anti-nonlinear periodic jamming

  • 期刊论文

  • 发表日期：

    2026年5月6日

查看全文
  • He Haitong,
  • Niu Yingtao

摘要

To address the challenges of slow convergence in large state-action spaces, limited utilization of prior knowledge, and constrained computational resources at terminal communication devices faced by reinforcement learning-based anti-jamming algorithms, this paper proposes a novel knowledge-assisted reinforcement learning algorithm for anti-nonlinear periodic jamming. The proposed algorithm enables communication receivers to sense channel jamming information to estimate whether dynamic jamming exhibits periodicity and to identify its period and pattern, thereby extracting anti-jamming knowledge. This knowledge is then embedded into the Q-table for initialization, with multi-value learning and pruning operations subsequently employed to further enhance the learning efficiency. Simulation results demonstrate that, under periodic dynamic jamming environments, the proposed knowledge-assisted multi-value reinforcement learning algorithm enables rapid response to evade jamming signals. Compared with conventional reinforcement learning methods without knowledge assistance, the proposed algorithm reduces the time required for effective anti-jamming decision-making by 90.9% to 93.9%, while achieving throughput improvements of 71.4% to 77.8% in dynamic jamming environments.