Attention-Enhanced DQN Scheduling for Multi-Link Devices in Synchronous N-STR Wi-Fi 7 Networks

WiFi 7 IEEE (802.11be) introduced Multi-Link Operation (MLO) that enables its devices to communicate over multiple links to support evolving latency-sensitive and high-throughput applications. However, MLO requires advanced scheduling algorithms to optimize the operation. This paper models WiFi 7 scheduling as a constrained Markov Decision Process that optimizes throughput, delay, latency and fairness while capturing access constraints and traffic dynamics. We also develop an attention-enhanced Rainbow Deep Q-Network (DQN) scheduling framework that combines multi-head attention, distributional Q-learning, and prioritized experience replay. Simulation results show up to 2.3x throughput improvement, 6x delay reduction, and marked gains in packet drop rate and spectral efficiency over baseline Round Robin scheduling.