Zero-Shot Generalization for Blockage Localization in mmWave Communication

This paper introduces a novel method for predicting blockages in millimeter-wave communication systems towards enabling reliable connectivity. Using a self-supervised learning approach, radio frequency data are labeled with blockage-causing object locations extracted from light detection and range data. These labeled RF data train a deep learning model that predicts object locations. The predicted positions are then used to forecast blockages, ensuring adaptability without retraining when the transmitter-receiver positions change. Experiments achieve up to 74% accuracy in dynamic environments, demonstrating the effectiveness of the proposed solution.