Reinforcement Learning-based Beam Hopping for Anti-jamming mmWave Communication

Farshad Soleiman, Massieh Kordi Boroujeny, Kai Zeng

Abstract: Beam hopping in millimeter-wave (mmWave) massive MIMO systems enhances resilience against jamming by dynamically switching transmission directions. However, optimizing beam hopping in real-time remains a challenge, particularly against adaptive jamming. This paper proposes a reinforcement learning (RL)-based beam hopping framework that intelligently selects beams to counteract static, round-robin, and random jamming. Simulation results show that our approach effectively mitigates interference, adapting to structured attacks while maintaining stable performance under unpredictable conditions. Additionally, we analyze the impact of jamming power, demonstrating that due to the sparsity of mmWave channels and highly directional beams, increased jammer power has a limited effect on SINR and throughput. These findings highlight the effectiveness of RL-driven beam hopping for securing future mmWave networks.