Abstract : The allocation of resources for bottleneck links in emergency communication networks, particularly under time-varying channel conditions, presents a critical challenge. Most existing studies predominantly focus exclusively on the independent optimization of either frame generation time or bandwidth allocation, overlooking the strong interdependence between these two factors. To address this gap, this paper proposes a joint optimization algorithm for frame generation time and bandwidth allocation (JFBA) based on a Double Deep Q-Network (DDQN). Our approach formulates the problem as a Markov Decision Process (MDP), where a DDQN-based agent perceives real-time network states to iteratively refine policies for both parameters. Furthermore, a tailored Prioritized Experience Replay (PER) mechanism, driven by both immediate rewards and Temporal-Difference (TD) errors, is incorporated to accelerate learning from rare but critical network anomalies, such as sudden traffic bursts or link failures. Extensive simulation results demonstrate that JFBA exhibits superior robustness compared to state-of-the-art baselines, including AFSA, DQN, PPO, and Dueling DQN. Specifically, under extreme resource constraints (e.g., 20% bandwidth availability), JFBA achieves an end-to-end delay reduction of up to 93.8% and maintains a bandwidth utilization rate exceeding 95%, successfully guaranteeing Quality of Service (QoS) where traditional methods fail to meet the constraints.
Index terms : Frame generation time, Bandwidth allocation, Joint optimization, Deep reinforcement learning, Prioritized experience replay