Enhancement of RL Approach Accessibility in NR

In wireless networking, research utilizing reinforcement learning is on the rise. In response, ns-3 has introduced additional modules like ns3-gym and ns3-ai to support such research. With the increasing traffic and diverse services, 5G NR has been developed and is supported in ns-3 via 5G-LENA. However, there's a need for more reinforcement learning examples based on NR. Additionally, improving the ease of use of multi-agent reinforcement learning (MARL) approaches aims to enhance the usability of NR Sidelink, enabling distributed operations. In this project, I will improve the interface between OpenGymEnv in ns3 and ns3env in the Python ns3 gym module for Multi-Agent Reinforcement Learning (MARL). Currently, ns3gym utilizes the REQ-REP pattern in ZeroMQ as the interface. By incorporating techniques such as client identity and parallel processing of workers in ZMQ, I aim to develop a MARL interface in ns3gym. Additionally, I will enhance the usability of 5G by implementing MARL approaches in 5G examples, with a specific focus on 5G LENA.