Geolocation with LoRa and machine learning using TDOA algorithm

Abstract

A vehicle tracking system is a crucial system used in fields such as transportation, delivery services, and criminal investigations. GPS and GSM system is the most popular means of determining vehicle location. It is accurate; however, it is costly and consumes a lot of power due to the need for frequent battery replacement. Implementing it in local public transportation, especially in African countries is not feasible. This is because installation and maintenance are expensive. Various research has been going on to improve geolocation with LoRa. However, one of the issues that remain is the time accuracy in collecting the time of arrival which is crucial in determining the vehicle position in the Time of Difference of Arrival (TDOA) algorithm. This project designed a low-cost and low-power LoRa vehicle tracking system and researched ways to improve LoRa geolocation using the TDOA algorithm using machine learning. A multilateration algorithm was used in positioning station gateways for the TDOA algorithm to provide an accurate measurement, it required a nanoseconds precision, however, only a millisecond precision was achieved from the hardware setup. The machine learning model used RandomForest for training and prediction advantages. The lack of a microcontroller with a nanoseconds precision, limited resources to build a full multilateration system affected the accuracy of the TDOA algorithm.