Brief Introduction

3GPP has completed the research on air interface based on AI/ML in the R18 phase and started the standardization of R19. Among them, data and models are the focus of research and standardization for R18 and R19. Based on simulated datasets, 3GPP has conducted extensive simulation verification work. Although AI models trained in simulated environments have shown good performance, the models used in different use cases in practical applications still need to be explored. Especially, for AI-based CSI compression, a bilateral model is needed, which is more difficult to implement compared to a single sided model, and a reference model is needed to provide guidance for the actual implementation and testing of the bilateral model.

IMT-2020(5G) Promotion Group-5G+AI Work Group has constructed public dataset platform and conducted reference model collection project to bridge the gap between theoretical research and real deployment. More than 40 companies and research institutes contribute to the public dataset and reference model. In this report, we focus on the data and models in 3GPP R18 and R19 study and standardization for AI-based CSI feedback compression. Reference models are provided for performance verification, test requirements determination and product realization. It is expected that these reference models could promote the integration of AI and wireless communication.