Discrete Positioning Using UWB Channel Impulse Responses and Machine Learning

Abstract

Automatic recognition of production tasks is a key aspect of an industrial Internet-of-things (IOT)environment. Often, the positions the tasks are executed at are of additional importance for process supervision and quality assurance. The complex structure of typical industrial environments including equipment, furniture and production objects leads to problems in obtaining the line-of-sight (LOS)connection necessary for precise localization with many RF-based systems. In this contribution, a method to obtain position estimates at a restricted set of points-of-interest via a machine-learning approach is proposed. The method is based on feature extraction on channel impulse responses (CIRs)of a Ultra-Wideband (UWB)radio system. It produces promising results in realistic scenarios, while the amount of data needed is small enough to enable retraining the database in a small amount of time. Additionally, the approach does not require calibration or synchronization of the UWB system and therefore could also be deployed in an existing system without additional configuration.