What's the Key Environment Information that Makes the Blockage Prediction Work?

Le résumé fourni par la source

To meet the increased communication performance requirements defined in sixth-generation (6G) networks, out-of-band methods are gaining more attention from researchers. These methods consider adding extra information, such as sensing environmental data, to enhance performance. In this context, image-based physical environment blockage prediction methods have been proposed to avoid fast signal attenuation, particularly in millimeter wave (mmWave) bands. However, image data is high-dimensional and requires specialized hardware for fast processing. This paper proposes a real-time blockage prediction method that utilizes images containing only propagation-related foreground information. The method predicts two blockage states using a convolution-based machine learning (ML) approach. We evaluate the proposed method using real-world communication data from a vehicle-to-everything (V2X) scenario. A comparative experiment is conducted to compare the performance of using complete images versus foreground-only images. The neural net-work's max pooling layer is configured with different parameters to test and analyze the impact of non-foreground information. The results show that the foreground-only image-based prediction is more robust for detail extraction and requires less prediction time. As the pooling stride increases, the accuracy of the complete image-based prediction method decreases, while the proposed method maintains high accuracy. Moreover, the proposed can save about 30 % of the prediction time.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.