A Pedestrian Indoor Navigation System Using Deep-Learning-Aided Cellular Signals and ZUPT-Aided Foot-Mounted IMUs

Abstract

A signal-sensor-based indoor pedestrian navigation system is developed. The proposed system: 1) utilizes a foot-mounted inertial navigation system (INS), in which the accumulated errors are mitigated via a zero velocity update (ZUPT) approach; 2) exploits opportunistically cellular long-term evolution (LTE) signals in a deep neural network (DNN)-based synthetic aperture navigation (SAN) framework, in which the pedestrian’s motion is utilized to suppress multipath-induced errors. The proposed DNN-SAN-LTE-ZUPT-INS (DUALS) indoor pedestrian navigation system utilizes the complementary the desirable characteristics of both subsystems, coupled via two architectures: (a) loosely-coupled and (b) tightly-coupled. This paper designs and assesses both architectures experimentally in an indoor environment. The experimental study demonstrates a pedestrian traversing a trajectory of 600 m in 14 minutes, including a stationary period, straight segments, up/down the stairs, and riding in an elevator, while receiving signals from 4 LTE base stations (also known as evolved node Bs (eNodeBs)). The proposed tightly-coupled DUALS system exhibited a three-dimensional (3-D) position root mean-squared error (RMSE) of 1.34 m, outperforming the loosely-coupled DUALS, ZUPT-aided INS, and LTE-DNN-SAN, which achieved a position RMSE of 1.38, 1.49, and 1.97 m, respectively.