D-TLoc: Deep Learning-aided Hybrid TDoA/AoA-based Localization

Beamforming with the multiple-input-multiple-output (MIMO) antenna arrays has been exploited to compensate significant signal power attenuation of high frequency wave. For proper communication, both the base station (BS) and the user equipment (UE) should transmit the beam into the optimal direction. In order to decide accurate beam direction and support beam management, localization techniques can be utilized; the conventional techniques are not adequate to handle the non-line-of-sight (NLoS) scenarios. An aim of this paper is to introduce the deep-learning aided uplink localization on the 3D plane. In this work, the deep neural network (DNN) is trained to estimate the the cartesian coordinate position of the device based on the information from uplink transmission, time difference of arrival (TDoA) and angle of arrival (AoA). Since the DL method is capable of skipping pre-processing stages, our proposed DNN is capable of accurately estimating the position of the device regardless of presence of NLoS paths. Using the dataset made from the raytracing simulation, we demonstrate that the proposed DNN predicts the position of 90% of the randomly placed devices within 0.2m.