Deep Reinforcement Learning based OFDMA Scheduling for WiFi Networks with Coexisting Latency-Sensitive and High-Throughput Services

In this paper, we focus on how to meet the diverse quality-of-service requirements of different types of STAs in WiFi networks where latency-sensitive (LS) STAs with strict end-to-end delay constraints coexist with high-throughput (HT) STAs demanding high data rates, which remains largely unexplored. To tackle this issue, we propose a novel deep reinforcement learning based resource scheduling algorithm for uplink Orthogonal Frequency Division Multiple Access (OFDMA) IEEE 802.11ax WiFi networks, aiming at guaranteeing the delay constraints for LS STAs while simultaneously maximizing the throughput of HT STAs. Extensive simulation results show that the proposed algorithm 1) exhibits substantial performance improvement compared to the traditional Round Robin algorithm under various traffic conditions, particularly for services with tighter delay constraints; 2) enables flexible resource allocation between LS and HT STAs by fine-tuning a predefined factor; and 3) achieves satisfactory reliability for LS STAs while good throughput for HT STAs under highly congested network conditions.