Session D1: Alternative Technologies for GNSS-Denied Environments

Precise Location Learning using Neighboring Cell IDs in LTE Networks: A Simple and Effective Approach
Jin Ah Kang, Youngsu Cho, Juil Jeon, Sun Sim Chun, and Jung Ho Lee, ETRI

In this paper, we introduce a deep learning-based text classification model as a novel method to improve cell-based positioning performance. Deep learning, which has attracted a lot of attention in the field of artificial intelligence, shows excellent performance in classifying text data. This is because deep learning methods automatically find important features from large amounts of text data, and this is possible because the text is divided into words or sentences and represented as vectors. By learning these words or sentences to be assigned meaningful places in a multidimensional space, deep learning models can learn the meaning of the text. Inspired by the deep learning-based text classification method, we devise a novel method to improve the performance of cell ID-based positioning. In other words, we represent the cell IDs of the serving cell and its neighboring cells collected from various locations in the mobile network as a vector space, and let the deep learning model discover the features important for positioning by itself. In this way, the deep learning model can learn the location using just raw data, namely cell IDs, instead of specific features such as the round-trip time (RTT) or the angle of arrival (AoA). To verify our approach, we first collected cellular measurements provided to a user equipment (UE) at various locations in long-term evolution (LTE) networks. Next, we analyzed the correlation between cell ID and precise location for fundamental verification, and then implemented a precise location deep learning model using long-short term memory (LSTM), similar to the text classification deep learning model. The experimental results showed that the proposed method can achieve precise positioning using only cell IDs.