Artificial Intelligence and Machine Learning Based Beam Prediction in 5G-Advanced and 6G Communications

Artificial intelligence (AI) or machine learning (ML) based spatial/temporal-domain beam prediction was studied by Release-18 and is currently being specified by Release-19, in the 3rd Generation Partnership Project (3GPP) fifth generation (5G)-Advanced standardization for future commercialization and standard evolution towards sixth generation (6G) communications. The goals for such 3GPP 5G-Advanced studies and standardization are to reduce user equipment (UE) power consumption by measuring less and to improve energy efficiency operation for network by transmitting less reference signal. In this paper, we first introduce major outcomes of Release-18 studies together with essential standardization components of Release-19 for AI/ML based beam prediction. To meet the intended goals and following Release-18/19 methodologies, we further investigate AI/ML based algorithms deployed at UE facilitating spatial and temporal beam prediction. Over-the-air performance evaluation based on 3GPP-compliant base station and UE with on-device AI/ML inference of the considered algorithms, are then further elaborated in this paper to demonstrate efficiency, accuracy, and feasibility of AI/ML based beam prediction with UE-side AI/ML. Finally, we present more recent experiments extending the Release-18/19 beam prediction schemes to mobility related use cases, which are promising to be considered for 6G standardization by 3GPP.