Deep Neural Network for Beam and Blockage Prediction in 3GPP-Based Indoor Hotspot Environments

Abstract

The application of millimeter-wave (mm-wave) frequencies in communication has the potential to address the ever-growing data traffic requirements of next-generation wireless communication devices. However, owing to their high directivity and high penetration loss, directional mm-wave beams are vulnerable to blockages caused by users' bodies and ambient obstacles. Further development of indoor mm-wave communication is essential, as the majority of data traffic is generated in indoor environments. In previous studies, the mm-wave blockage problem was primarily considered in outdoor scenarios, whereas in the present study, online learning-based beam and blockage prediction in an indoor hotspot (InH) scenario was investigated. During an offline training phase, we constructed a fingerprinting database consisting of user locations along with their respective data traffic demands and corresponding blockage statuses with optimal beam indices. Following its creation, the fingerprinting database was used to train the weights and bias of a properly designed deep neural network (DNN). During a subsequent online learning phase, the trained DNN was fed user locations and corresponding data traffic demands at the served user equipment to output optimal beam indices and blockage statuses. System-level simulations were conducted to assess the accuracy of blockage prediction based on 3GPP's new radio channel and blockage models in InH environments. Simulation results revealed that the proposed scheme was capable of predicting mm-wave blockages with an accuracy of > 90%. These results confirmed the viability of the proposed DNN model for predicting optimal mm-wave beams and spectral efficiencies.