Key points are not available for this paper at this time.
The paper describes possible deep reinforcement learning algorithm for beam tracking and prediction of user equipment angular motion. High accuracy in predicting azimuth and elevation angles is achieved by applying an advanced learning method - Proximal Policy Optimization. To track the time-varying sequence of angular coordinates more accurately, Long-Short Term Memory cells are used as one of the implementations of recurrent neural networks. The proposed approach does not require any a priori knowledge about environment and can be implemented in real time.
Averina et al. (Wed,) studied this question.