A Federated Hierarchical DQN-Based Distributed Intelligent Anti-Jamming Method for UAVs

Area of Science:

• Intelligent communication systems
• Deep learning applications in UAVs
• Cooperative decision-making for drone swarms

Background:

• Deep learning anti-jamming is crucial for unmanned aerial vehicle (UAV) systems.
• Single-UAV models face data isolation and decision inconsistencies in swarms.
• Data sharing in swarms increases communication overhead and security risks.

Purpose of the Study:

• To propose a novel multi-UAV cooperative intelligent anti-jamming decision-making method.
• To address data isolation, decision inconsistency, communication overhead, and security challenges.
• To enhance the effectiveness and efficiency of anti-jamming strategies in UAV swarms.

Main Methods:

• Federated Learning-Hierarchical Deep Q-Network (FL-HDQN) framework.
• Adaptive model synchronization for collaborative global model training using local parameters.
• Hierarchical deep reinforcement learning for multi-domain optimization (time-frequency, power, modulation-coding).

Main Results:

• FL-HDQN ensures decision consistency and preserves data privacy by sharing model parameters, not raw data.
• The hierarchical model effectively decouples complex optimization into manageable layers.
• Achieved 1% higher decision accuracy compared to state-of-the-art intelligent anti-jamming models.

Conclusions:

• The proposed FL-HDQN method offers a superior and effective solution for cooperative intelligent anti-jamming in UAV swarms.
• Federated learning significantly reduces communication costs and enhances data security.
• Hierarchical deep reinforcement learning improves decision-making performance in complex interference environments.