Deep Learning-Based Proactive Physical Layer Handover using Cameras for Indoor Environment

Camera-assisted prediction-based handover has been implemented in a proactive manner for an indoor dynamic 60 GHz radio channel. Here, our proposed deep learning model is utilized to forecast the decline in signal quality caused by blockage. The predictive model shows equivalent accuracy and faster training time in comparison with models using state-of-the-art deep learning architectures of ResNet-18 and ResNet-50, owing to its suitability to the data obtained in the proposed handover experiment. Accordingly, a proactive physical layer handover method is proposed, which is based on the link quality prediction. This method outperforms handover in a reactive man-ner, as evidenced by longer connected duration, demonstrating its feasibility of malntaining a seamless connection.