Indoor Height Recovery by Cellular Signal Data: Poster Abstract

Understanding fine-grained distribution of telecommunication (Telco) indoor signals within high buildings is important for Telco operators to optimize wireless communication quality. A key task is to tag Telco indoor signals with associated height labels, i.e., the so-called indoor height recovery problem. Unlike existing works requiring sufficient training data within a single building, it is rather hard to generalize the problem across city-scale buildings, especially for those target buildings with scarce training data. To this end, we consider how to perform the height recovery problem for such target buildings with help of source buildings having sufficient labelled Telco signal data. We present a deep neural network (DNN)-based framework which involves three key steps (Telco signal parametrization, spatial image construction and DNN-based height estimation) to fill the gap between the buildings. Our preliminary evaluation demonstrates the potential of our work.