Low Sweeping Overhead Method Based on Machine Learning in Beam Selection

Massive multi-antenna technology enhances spectral efficiency and system capacity in multi-user scenarios, thereby fulfilling the escalating demand for superior communication quality. This technology generates an increased number of spatial channels, referred to as independent beam pairs, to achieve higher spatial multiplexing. In such scenarios, users may select the optimal beam pair based on the conditions of their channels. The plethora of beams produced by a massive multi-antenna system presents a significant challenge in swiftly and efficiently selecting the optimal beam pair during initial connection setups. This paper proposes a method based on deep neural networks (DNN) that reduces the number of beam pairs needing consideration during the selection process, thereby minimizing overhead by searching for the optimal beam pair among a few recommended by the neural network.