A Modular framework for AI/ML applied to B5G V2X networks co-simulations

Abstract

The use of AI/Ml is a key feature for the new generations of mobile communications as the actual generations are becoming more complex in order to provide a faster connection for a large number of users, but in general, it requires large datasets in order to produce high-quality AI/ML models. Due to the costs of collecting real measurements, especially in vehicular and aerial scenarios, this dissertation proposes a modular methodology that enables the combination of different simulators in order to produce realistic datasets for V2X and aerial cellular communications. The methodology also brings the possibility to train AI/ML models in-loop with the simulations. Furthermore, this dissertation details benchmarks for CPU and simulation time in different simulation scenarios and also the results of a use case showing the data that can be extracted from the combinations of the used simulators, where it is possible to observe that the use of parallel computing can reduce the simulation time by approximately five times.