Deep Reinforcement Learning-Based Fairness and Throughput-Aware Association Control Algorithm for Dense WLAN Systems

Dense indoor wireless local area networks (WLANs) suffer from load imbalance and performance degradation due to independent and uncoordinated Wi-Fi access point (AP) operations. Although existing association-control schemes have improved throughput, fairness, or quality of experience (QoE), these schemes often rely on infrastructure-side modification or fail to jointly consider system throughput and fairness in user demand satisfaction. To address these limitations, we introduce a user-centric software-defined WLAN (SD-WLAN) architecture that enables centralized association control without modifying legacy AP infrastructure. In addition, we propose a deep reinforcement learning-based fairness- and throughput-aware association-control (Deep-FTAC) algorithm. Deep-FTAC employs the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm together with a Straight-Through Gumbel-Softmax (STGS) mechanism to support differentiable discrete AP selection. Simulation results demonstrate that Deep-FTAC improves the overall system throughput and user demand-satisfaction fairness by up to 14% and 41%, respectively, compared to the conventional scheme in which each user is associated with the geographically closest feasible AP.