Non-Line-of-Sight Identification for UWB Indoor Positioning Systems using Support Vector Machines

This paper presents a Non-Line-Of-Sight (NLOS) identification approach based on machine learning algorithms for ultra wide band positioning systems. The identification of NLOS conditions is crucial for positioning using trilateration as NLOS introduces positive biases in the calculated distances. The proposed method is based on the classification of the Channel Impulse Responses using Fisher's Linear Discriminant and Support Vector Machines (SVM). The proposed approach has been validated by measurements in both an anechoic chamber where known reflections and obstacles are introduced and in a basement corridor as real environment scenario with more than 500 and 700 measured data sets for training, respectively. Results show an average identification accuracy of 92% for the case using SVM in the anechoic chamber and almost 100% for Fisher's discriminant combined with SVM for the corridor scenario.

I. Introduction

Indoor wireless positioning systems offer a complementary option to satellite based positioning systems when it is difficult to receive signals from satellites in an indoor environments. Typical application examples could be autonomous robots in a warehouse or self-driving cars in underground parking lots. Ultra Wide Band (UWB) uses Time Of Flight (TOF) to obtain the distance between a tag and an anchor. If no clear path exists between the tag and the anchor i.e. the received signal would travel a greater distance between the tag and anchor than there actually exists [1]. This situation is known as Non-Line-of-Sight (NLOS). This positive bias can lead to big positioning errors and cause severe consequences such as car accidents.