Publication
TraPos: A Transformer-Based Localization Framework for Next-Generation Mobile Networks
Till Ruppert; Florian Langenstein; Dennis Salzmann; Florian Herrmann; Christoph Fischer; Michael Gundall; Hans Dieter Schotten (Hrsg.)
IEEE International Conference on Communications (ICC-2026), IEEE, 2026.
Abstract
In multipath-rich indoor environments, accurate
localization remains a challenging task, as bandwidth-limited
signals require high-resolution algorithms for precise parameter
estimation. To address this challenge, we introduce TraPos,
a transformer-based framework for indoor localization. The
network is trained and evaluated on 5G channel measurements
collected with a software-defined radio (SDR) platform. The
datasets were acquired using an automated guided vehicle
(AGV), with spatial uniformity ensured through grid-based
filtering. As a baseline, TraPos is compared to a state-of-the-
art joint angle-delay Multiple Signal Classification (MUSIC)
algorithm. Results show that TraPos outperforms the super-
resolution method in both angle of arrival (AOA) and time
difference of arrival (TDOA) estimation. When trained directly
on two-dimensional coordinates, TraPos achieves decimeter-level
accuracy, demonstrating competitive performance for unfiltered
position estimates.
Projects
- Open6GHub (2021) - 6G for Society and Sustainability
- Open6GHub+ - Offenes 6G Transferzentrum für Wirtschaft, Wissenschaft und Gesellschaft