Beam prediction and tracking mechanism with enhanced LSTM for mmWave aerial base station

By combining millimeter wave (mmWave) with abundant spectral resources and advanced directional beamforming technology, mmWave aerial base stations (mAeBSs) can provide high-speed services to users on the ground while reducing interference to existing terrestrial networks. Due to the mobility of users and the corresponding changes of other environmental factors, the beam misalignment between mAeBSs and users will pose a serious challenge to the stability and quality of communication link. Consequently, we propose schemes based on deep learning for beam prediction tracking in this paper. On the basis of received signals from both the present and previous beam training, we apply long short term memory (LSTM) module to realize the beam prediction. In order to enhance performance of LSTM scheme, cascaded LSTMs are designed in the model. Furthermore, we attain the derivative of implicit state function of the beam changes to formulate an ordinary differential equation (ODE) problem and construct the LSTM_ODE model for beam prediction. Simulations demonstrate that, in comparison to existing other schemes, our proposed approaches attain greater beamforming gain while incurring less overhead during beam training.