Deep Learning-Based Blockage Prediction For Intelligent Reflecting Surfaces-Aided V2X Networks

We consider the problem of prediction of channel blockage in an intelligent reflecting surfaces (IRS)-assisted vehicle-to-everything (V2X) network, using deep learning (DL) techniques. The vehicular communication is assumed to follow millimeter-wave technology, and the straight link between the transmitter and receiver is presumed to be unavailable. For the prediction problem, we consider three DL architectures, namely, the convolutional neural network (CNN), long shortterm memory (LSTM) and gate recurrent unit (GRU) networks. In particular, we show that using only a minute portion of overall elements in IRS to be active results in an increase in SNR and prediction accuracy. Our detailed experiments on the visionaided wireless communications (ViWi) dataset show that LSTM offers a good overall average prediction accuracy and least root mean squared error performance, compared to CNN and GRU.