Vision-Aided Blockage Prediction and Proactive Handover for Indoor Terahertz Communications

Abstract

To support extremely high data rates in 6G wireless networks, Terahertz (Thz) communications that explore the abundant spectrum resources at the Thz band have attracted great interest in recent years. However, due to the strong directivity and se- vere signal attenuation of THz signals, the link quality is highly sensitive to obstacles, especially when there is only a line-of-sight (LoS) path. To enable proactive handover to a transmitter with an alternative LoS link, accurate blockage prediction is essential to avoid the sudden drop in transmission quality. Unfortunately, existing methods fo- cusing on outdoor environments often fail to predict blockages in complicated indoor environments.

In this paper, we propose a background-aware vision-aided blockage prediction framework that utilizes the sequences of historical RGB-depth (RGB-D) information and the beam indices to detect and localize users and potential blockages, predict their trajectories, and foresee the blockages in dynamic indoor scenarios. Specifically, we first model the background with the medium filter and use a deep-learning-based object detector to detect the users as well as potential blockages. We then predict the future locations of the users using an LSTM-based neural network and predict the time when the users locate behind the background. We demonstrate from numerical results that the proposed scheme outperforms conventional schemes in terms of blockage predic- tion and proactive handover decision-making accuracy.