The sensitivity of millimeter-wave (mmWave) to blockage and the requirement for the communication system to support mobility scenarios makes mmWave blockage and beam prediction necessary. In this paper, the effect of diffraction characteristics on improving blockage and beam prediction is investigated. A dataset with diffraction is created, and a recurrent neural network (RNN) is designed to capture the diffraction characteristics in the dataset. Further, the generalization ability of the RNN for different scenarios is researched. The results show that the accuracy of both blockage and beam prediction is further improved by utilizing diffraction characteristics. In addition, the blockage and beam prediction accuracy of the designed RNN is increased by 1.5% and 9.7% compared with a reference deep neural networks (DNN). Finally, applying the RNN model trained in the outdoor scenario to the prediction in the indoor scenario, the indoor blockage and beam prediction accuracy respectively reach up to 99.8% and 98.3% of the outdoor’s.
Diffraction Characteristics Aided Blockage and Beam Prediction for mmWave Communications
01.06.2022
3489749 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch