Collaborative Multi-Domain Anti-Jamming for Low-Altitude Economy Networks: An Attention-Enhanced MARL Approach

Low-Altitude Economy (LAE) networks have emerged as a critical enabler for cooperative aerial missions such as urban logistics, emergency rescue, and aerial mobility. However, due to the exposed electromagnetic environment, wireless links in LAE networks are inherently vulnerable to jamming attacks. In this paper, we focus on dense LAE networks and investigate a distributed multi-node anti-jamming strategy for the network. We first formulate a multi-agent Partially Observable Markov Decision Process (POMDP) to achieve multi-domain synergy by jointly optimizing frequency hopping, transmit power, and Modulation and Coding Scheme (MCS). To derive a collaborative solution without a central controller, we then propose a novel Multi-Agent Reinforcement Learning (MARL) algorithm by adopting an agent-state-aware attention mechanism, which dynamically re-weights the agent importance in the learning process. Moreover, we prove the existence of an optimal stationary deterministic joint policy for the proposed problem, and analyze the convergence stability. Finally, extensive simulations demonstrate that the proposed algorithm significantly outperforms baselines in terms of effective data rate and Packet Delivery Ratio (PDR), improving the effective data rate by approximately 10%, while simultaneously reducing synchronization overhead and maintaining a superior transmission rate.