Spectrum and Device State-Aware Anti-Jamming Resource Allocation for Dynamic Uav Networks

Securing communication links against malicious jamming has become critical with the widespread deployment of mobile unmanned aerial vehicle (UAV) networks. However, most existing UAV network anti-jamming methods rely solely on spectrum observations, overlooking the channel aging phenomenon induced by high mobility of UAVs. This paper presents a new joint spectrum and power allocation anti-jamming scheme based on the spectrum and device state-aware dueling double deep Qnetwork (SDSA-DDQN) for dynamic UAV networks. First, we construct a high-fidelity dynamic UAV environment model to simulate realistic non-linear link outages by incorporating the channel aging mechanism and a Sigmoid-based reliability factor. Using this model, the anti-jamming resource allocation problem is formulated as a Stackelberg game. To extract multi-modal features, we design a hybrid dueling network utilizing a parallel convolutional neural network (CNN) and multi-layer perceptron (MLP) architecture. We fuse spectrum observations with device mobility states and thus, the agent can effectively decouple aging-induced errors from interference. Finally, we integrate the double DQN framework with the prioritized experience replay (PER) mechanism. As a result, value overestimation can be mitigated and training convergence is significantly accelerated. Simulation results demonstrate that our scheme offers significant advantages over the state of the art in terms of convergence speed and system utility.