A Comparative Performance of Wi-Fi Spatial Reuse Solutions in 802.11ax

This paper presents a performance comparison of six spatial reuse schemes for IEEE 802.11ax (Wi-Fi 6), ranging from centralized and distributed designs and from reinforcement-learning and heuristic approaches. Each algorithm is implemented in the ns3 simulation environment and assessed under a unified methodology: single 5 GHz channel, 20 MHz bandwidth, fixed MCS4, saturated UDP downlink, one-to-one AP–STA service per slot, and BSS Coloring/OBSS-PD enabled by default. Three topologies with various node densities were used, and the algorithm performance is evaluated using the aggregate throughput and Jain’s fairness metrics. Simulation results show that when channel selection is disallowed, transmit power control is the dominant lever for spatial reuse, while OBSS-PD and contention window (CW) tuning provide limited throughput and fairness optimization gains. OBSS-PD by itself can degrade fairness in poorer channel conditions without improving throughput, indicating it should operate jointly with power/CW adaptation. In addition, amongst several popular reinforcement learning (RL) based methods that were evaluated, a centralized multi-armed-bandit controller consistently achieves the best throughput–fairness trade-off, while the distributed RL based method (e.g., DQN/DDPG variants) is sensitive to local observability and tends to converge to suboptimal yet stable operating points in static topologies. Overall, careful power adaptation guided by centralized coordination, is the most effective for Wi-Fi 6 spatial reuse in our tested topologies conditions.