Predicting Power Density for Mm-Wave Handset Antennas Based on Machine Learning

Abstract

Simulating the peak spatial-average incident power density (PD) for 5G millimeter-wave (mm-wave) handset antenna arrays is important but time-consuming, as the entire handset needs to be taken into account. In this letter, a time-efficient method based on machine learning (ML) is proposed to predict the maximum PD of mm-wave antennas in real handsets. By training various typical antenna arrays, a mapping relationship is established between the near-field distributions of the antenna array in real handsets and those of the stand-alone array. With this mapping, the maximum PD of any mm-wave array in a real handset can be quickly predicted by simulating the PD of the corresponding stand-alone array. The method has been verified on several 4 × 1 arrays commonly used in 5G mobile handsets, showing accurate prediction of the maximum PD. The proposed method can be applied to mm-wave arrays of any type and significantly reduces the evaluation time.