Artificial Intelligence-Based Beam Management for High Speed Applications in mmWave Spectrum

Future new radio (NR) systems will support high speed applications in the millimeter wave (mmWave) range. To improve user experience, accurate beam management is proposed based on artificial intelligence (AI) technology. This paper also proposes a training set generation method and its manipulation which can properly train the proposed algorithm by exploiting channel characteristics. To collect measurement data before training, we propose a practical data acquisition method based on the reference signal defined in the current NR specification. Evaluation results show that the accuracy approaches 98 percent for beam selection both with ideal hardware and with hardware impairments including phase noise and amplitude noise, as the number of epoch increases. From the results, we confirm that the proposed AI-based beam management can predict the desired beam accurately as well as has robustness to hardware impairments that are crucial in millimeter wave spectrum.