Performance dependency of LSTM and NAR beamformers with respect to sensor array properties in millimeter‐wave V2I scenario

Abstract

Interference prediction is a challenging problem in millimeter-wave V2I scenarios. The implementation of a practical V2I network is limited because of the interference due to the random nature of the wireless channel. This paper proposes an adaptive beamforming technique for mitigation of interference in V2I networks. In this work, long short-term memory (LSTM) and nonlinear autoregressive (NAR)-based regressors have been employed to predict the angles between the RSU and UE. Advance prediction of transmit and receive signals enables reliable V2I communication. Instead of predicting the beamforming matrix directly, we predict the main features using LSTM for learning dependencies in the input time series where complex variables were taken as input states and the final beamformed signal was the output.Simulation results have confirmed that the proposed LSTM model achieves comparable performance in terms of system throughput compared to the NAR method implemented as an artificial neural network.