Transfer Learning of Wi-Fi FTM Responder Positioning with NLOS Identification

Abstract

This paper proposes a neural network (NN) model-based method to predict the location of Wi-Fi access points (AP) that support the fine time measurement (FTM) protocol. From the collected FTM data by the robot, the NN is trained on one FTM responder (FTMR) data to recognize non-line-of-sight (NLOS) patterns. Even without knowing the offsets of individual FTMRs in advance, the model can be used to predict the locations of those FTMRs. The knowledge of FTMR location can be further used to understand network density, connectivity and interference characteristics in buildings. Compared with the results of the basic least-squares and circular positioning method, the experimental results increase the positioning accuracy by 86% and 83%, respectively.