Using Public Datasets to Train O-RAN Deep Learning Models

The disaggregation promoted by the O-RAN architecture allows for unprecedented flexibility in Radio Access Networks (RANs). The existence of specific components to control the infrastructure, such as RAN Intelligent Controllers (RICs), places intelligence at the center of the management and orchestration mechanisms of these networks. Hence, deep learning plays a crucial role in developing these solutions. As deep learning heavily relies on data for model training and generalization, using public datasets becomes essential to facilitate research and foster advancements in O-RAN. This paper surveys the primary public datasets available online used in O-RAN research. Our goal is to overview these datasets and act as a complement to their documentation.