Deep Reinforcement Learning Based Spatial Reuse for IEEE 802.11 bn

Recently, the Project Authorization Request (PAR) for IEEE 802.11 bn (Wi-Fi 8) has outlined throughput, the 95th percentile of latency distribution, and medium access control Protocol Data Unit loss as pivotal optimization indicators. In the meanwhile, Spatial Reuse (SR) is harnessed to facilitate concurrent transmissions, aiming to establish highly efficient wireless local area networks. In this paper, we introduce a deep reinforcement learning based approach to optimize SR operation. Taking Wi-Fi 8 PAR into account, we mathematically model the optimization problem and reward function is exquisitely designed to align with the optimization objectives. Our proposed method incorporates a novel Markov state transition process, accounting for information from other nodes in both state and reward. Each node is empowered to autonomously decide whether to transmit, with simultaneous implementation of rate adaptation and power control. We evaluate the performance of our algorithm in random topologies and dynamic environments, demonstrating significant advancements in both latency reduction and loss mitigation.