Towards 6G: Beam Prediction using Convolutional Neural Network and Artificial Neural Network

The sixth generation (6G) mobile network will be one of the revolutionary technologies in the next decade. The 6G will use mmWave and terahertz (THz) waves. However, mmWave and THz communication have higher path loss and absorption loss. To overcome this, beamforming (BF) is used. Therefore, beam training methods are needed to precisely align the transmitter and receiver. However, beam training requires more computation. Recently, research have shown significant interest in beam prediction using machine learning. This paper introduces an approach that utilizes machine learning techniques along with data-cleaning methods for beam prediction. The model contains a convolutional neural network (CNN) and a multilayer perceptron (MLP) network that accepts two types of inputs which are images and global positioning system (GPS) data. This study used an available online dataset for the evaluation of the model. The dataset comprises GPS data and images accompanying other data like beam indexes, received power, etc. This study used a portion of this dataset that consisted of 15281 images and 15281 GPS data samples. The study tested the proposed approach with 64 and 32 beam indexes which are output classes. The model achieved 88.95% and 82.23% accuracies for 64 and 32 classification classes respectively.