Gymir5G: A Simulation Platform to Study Data Transmission over WebRTC in 5G Networks with Deep Learning Assistance

This paper presents Gymir5G, an open-source simulation platform for running various simulated scenarios in 4/5G and $\mathrm{Wi}-\mathrm{Fi}$ networks. It is primarily designed to study the optimization of real-time multimedia data transmission in cellular networks using deep learning algorithms, later to apply the resulting trained models on real remote-controlled devices. Based on the INET and Simu5G libraries from the OMNeT++ discrete event simulator ecosystem written in C++, Gymir5G introduces three main features. First, it presents a basic version of the WebRTC protocol with all of its key congestion control, retransmission, and redundancy mechanisms implemented. Second, there is a separate Python project, GymirDRL, which introduces a deep reinforcement learning setup as an example of a deep learning method applied to data optimization, and an inter-process communication bridge to integrate simulation and learning parts under one solution. Gymir5G provides realistic base station parameters, configurable data streams, and different scenarios including non-standalone and standalone 5G networks, as well as multihoming configurations for the parallel use of cellular and wireless stacks. Finally, Gymir5G presents two example simulations, one of which is designed for the real geographic area in the Bay of Kiel, Germany, and the moving ship as a remote-controlled device.