Dual Intelligence: Leveraging DRL With Smart Satellites to Counter Intelligent Jamming in Satellite Networks

Satellite networks provide a powerful solution for reliable wide-area coverage and robust communication in diverse environments. However, open ground-to-satellite links are highly vulnerable to environmental dynamics and interference, especially the prevalent intelligent jamming attack. To address these challenges, we develop a dual intelligence framework in satellite networks that enhances the random access (RA) efficiency by incorporating intelligence at both the user and satellite sides. On the user terminal (UT) side, we develop an error-attention deep deterministic policy gradient (EADDPG) algorithm that prioritizes samples with larger time-difference (TD) errors, to enhance the training efficiency and accelerate convergence in dynamic environments. On the satellite side, as environmental instability increases, the jamming effectiveness of the intelligent jammer intensifies. Therefore, we introduce the intelligent satellite model to countermeasure the effectiveness increase of jammers in dynamic environments by sending false negative acknowledgment (NACK) messages to disturb the learning process. Simulation results show that the proposed method can effectively mitigate the performance degradation caused by dynamic environments and effectively counter intelligent jamming. The EADDPG outperforms other vanilla deep reinforcement learning (DRL) algorithms in anti-jamming capability, with interference occurrences reduced by nearly 50% across various environments compared to vanilla deep reinforcement learning algorithms.