This project is part of our academic curriculum in the third year of the engineering cycle. The objective of this work is to put into practice the theory that we have studied throughout the year.
mmWaveTo enable highly mobile millimeter wave (mmWave) and terahertz (THz) wireless communication applications, key challenges associated with the large antenna arrays deployed in these systems must be overcome. especially custom, The narrow beams of these antenna arrays typically result in high beam training proportional to the number of antennas. To address these challenges, our project proposes a multimodal hybrid deep learning and machine learning approach for fast beam prediction using position (GPS) and vision (camera) data from a wireless communication environment.
The developed framework is tested on a real vehicle dataset, which includes practical GPS and cameras. We study position-based Beam prediction models using MLP (Multilayer Perceptron), KNN (K-nearest neighbors) algorithms, and vision based Beam prediction models using ResNet-50. Then, we experimented with different data fusion techniques : pre- (Early Fusion) and post- (Late Fusion) models. The results show that the proposed fusion methods achieves more than 87% Top-3 beam prediction accuracy in realistic communication scenarios.