AI Pseudo-Cell: A Method for Generating MultiCell Measurements for Precise Positioning Using Language Translation Models

Le résumé fourni par la source

This paper proposes a novel and practical method, termed artificial intelligence (AI) Pseudo-Cell, designed to enhance the accuracy of cell-based positioning in mobile networks. The approach augments conventional cell measurements (CMs), which are commonly used in existing cell-based positioning methods. Specifically, AI Pseudo-Cell employs neural machine translation (NMT) models— typically used in natural language processing—to generate neighboring CMs from other mobile network operators (MNOs), using the CMs of the subscribed MNO connected to a user equipment (UE) as input. The generated multi-MNO CMs can be effectively utilized to enhance the positioning precision compared to using a single-MNO’s CMs. To implement AI Pseudo-Cell, we collected multi-MNO CMs from diverse geographic locations using a custom-built device equipped with three long term evolution (LTE) modems, each subscribed to a different MNO. Based on the collected data, multiple NMT models were trained to translate CMs between MNOs at the same location. Experimental results demonstrate that the AI Pseudo-Cell model can generate CMs from other MNOs with an average accuracy over 71% across training, evaluation, and indoor datasets, while achieving an average cell ID match rate of 86%. These findings suggest that the AI Pseudo-Cell can enable precise location inference even in challenging environments such as global navigation satellite system (GNSS) shadow zones.

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