Improvement of accuracy of UWB Positioning System within the intersection using a Long Short-Term Memory Network

In this paper the pedestrian positioning system in the intersection using UWB(Ultra-Wide Band) is studied. In this system, UWB signal is transmitted from the terminal owned by each pedestrian and it is received by base stations attached to traffic lights for pedestrian, and the distance to the pedestrian is measured. Distance between the pedestrian and the base station is estimated by TOA (Time Of Arrival) of the received signal. The position of the pedestrian is estimated by LSM (Least Square Method) using the distance measurement value. The accuracy of UWB positioning system deteriorates due to thermal noise and multipath. The distribution of this accuracy differs for each position, and it is complicated to create a mathematical model to correct. In this paper, appropriate correction is performed by learning different noise distributions for each position using LSTM (Long Short-Term Memory), a kind of RNN (Recurrent Neural Network). In addition, we propose a network that estimates the current value using previous observations in order to estimate the position even if the positioning of the pedestrian fails. The effectiveness of the proposed method is verified by comparing the performance with positioning using the Kalman filter. As a result of computer simulation, it is shown that the proposed method achieves high accuracy positioning performance.