AI-Based Interference Pattern Prediction and Fair Scheduling in Full-Duplex Wireless IoT Networks

Full-Duplex (FD) wireless is an emerging technology that enables a wireless device to transmit and receive at the same time and on the same frequency band. This can significantly increase the total number of IoT devices that a single access point (AP) can serve. However, in networks with half-duplex (HD) clients, FD operation at the AP can introduce strong interference from uplink (UL) clients to downlink (DL) clients, limiting the system scalability. Efficient scheduling (or pairing) of UL and DL clients can mitigate this interference by selecting clients with low mutual interference (i.e., high path loss). However, acquiring instantaneous channel information between clients to make such scheduling decisions can result in substantial overhead, potentially eliminating all FD capacity gains. This paper introduces a system called PF-FD to address these problems and achieve fairness in multi-AP WiFi-based IoT networks. At its core, PF-FD employs (i) machine learning techniques so that each AP can autonomously predict path loss (or interference relationships) between devices with a high accuracy and low overhead; (ii) A linear complexity scheduling algorithm that helps each AP make scheduling decisions based on its path loss estimates, with outputs that are very close to the proportional fair outcome; and (iii) A distributed contention-based medium access control (MAC) and rate adaptation protocol compatible with legacy WiFi. We evaluate PF-FD with Python and NS3 and show that its path loss estimation method has a very high resiliency to missing data and uncertainties from environmental factors, and that it provides up to 21% more throughput than state-of-the-art.