Improving TDoA Based Positioning Accuracy Using Machine Learning in a LoRaWan Environment

Abstract

LoRa is one of the low power wide area communication technologies (LPWA) that enables low cost chip module design due to low power, high receiver sensitivity and license-exempt bandwidth. Because of this, It is a technology suitable for IoT services with low data throughput and variability. For low-power-based positioning in $L$ oRa environments While varinous techniques have been tried, The error is It is over a hundred meters. Because of this It is difficult to commercialize practical location services. In this paper, To reduce the TDoA positioning error, a train was made to correct the time error that occurs when transmitting. We propose a method of learning the time error in the DNN model and correcting it using the learned model in actual positioning. The experimental environment was constructed using python and keras. Experiment result, We confirmed that the error range decreases when the number of reference nodes and collected data are large and the mobile node is close to the reference node.