A Mobile-Centric Approach to Improving the Performance of Decentralized Wireless Networks

Abstract

The proliferation of wireless devices and intelligent environments has led to increasingly dense, interference-prone Wireless Local Area Network (WLAN) deployments. Recent amendments, such as IEEE 802.11ax/be, aim to enhance spatial reuse (SR) by optimizing spectrum efficiency. However, current SR initiatives face significant deployment challenges, including incompatibility with legacy 802.11 versions, a lack of centralized coordination, the difficulty for Access Points (APs) to monitor their neighborhood while maintaining uninterrupted service, and limited monitoring range. We argue that these limitations stem from the traditional AP-centric approach. In contrast, this work proposes a mobile-centric SR approach that overcomes these issues. Our method uses roaming mobile terminals (MTs) to independently monitor neighboring AP parameters (channel, transmission power, carrier-sense threshold), regardless of the 802.11 version or owner. After collecting local topological information, MTs execute a two-stage optimization: first, a local channel selection heuristic, followed by an algorithm that uses a pre-trained neural network to quickly evaluate and select the best transmission power and carrier-sense threshold for each in-range AP. MTs then broadcast these recommended parameters opportunistically using unassociated management frames, allowing them to be modified on the fly. By using a pre-trained model, our system does not need to learn in real time; consequently, network convergence is primarily governed by how quickly the mobile terminals explore and cover the deployment area, which depends on factors such as terminal speed, number of mobile terminals, trajectory characteristics, and pause times. Simulation results demonstrate that this mobile-centric approach quickly identifies near-optimal AP configurations, achieving throughput performance comparable to the reinforcement learning approaches considered in this study while exhibiting substantially shorter estimate

The authors' abstract, as published at the source. Information, 2026 · DOI ↗