Deep Reinforcement Learning for Access Traffic Splitting in 5G Core System

The integration of 3GPP Radio Access Technology (RAT) (LTE, 5G New Radio (NR)) with non-3GPP RAT (WiFi) was highlighted as one of the key innovations of 5G technology by the International Telecommunication Union (ITU) in the IMT-2020 (International Mobile Telecommunications) standards. This integration ensures service quality and user experience even during congestion in mobile networks due to the increasing number of devices. Based on the standards provided by 3GPP regarding Access Traffic Steering, Switching, and Splitting (ATSSS), we have developed a deep reinforcement learning-based agent to integrate into the Policy Control Function (PCF) of the 5G Core System. This agent is capable of making guidance for offloading NR traffic to WiFi in case of NR network congestion. The agent outperforms traditional methods by ensuring the latency requirements for end-users while maximizing the revenue for the network operator.