An Effective Tree-Structured AI Model for Reducing Overhead of Life Cycle Management in Wireless Communication

Abstract

Artificial intelligence (AI) has been widely applied across various industries, including wireless communication. AI has been a topic of extensive discussion within the 3rd Generation Partnership Project (3GPP), particularly in the context of the physical layer. It involves applications such as beam management, positioning accuracy enhancement, and channel state information (CSI) feedback improvement. Evaluation results from various companies indicate significant gains in beam management and positioning through AI integration. While AI can replace traditional mechanisms to enhance performance, it also introduces new overheads. For example, the introduction of AI models necessitates addressing model life cycle management (LCM) issues, such as model identification, activation/deactivation, monitoring, updating/finetuning, selection, switching. These operations result in a significant amount of overhead. In this paper, we present the progress of model LCM in 3GPP, and also propose a tree-structured model to reduce the overhead associated with LCM operations. This model can be used in scenarios where multiple models serve the same functionality by merging similar structures, thereby saving storage space, and making the processes of model switching, expansion, and deletion more effective. We also conduct simulations to demonstrate that our approach maintains stable AI model performance while simplifying the model structure.