Collision prediction for a low power wide area network using deep learning methods
Shengmin Cui, Inwhee Joe
Abstract
A low power wide area network (LPWAN) is becoming a popular technology since more and more industrial Internet of things (IoT) applications rely on it. It is able to provide long distance wireless communication with great power saving. Given the fact that an LPWAN covers a wide area where all end nodes communicate directly to a few gateways, a large number of devices have to share the gateway. In this situation, chances are many collisions could occur, leading to waste of limited wireless resources. However, many factors affecting the number of collisions that cannot be solved by traditional time series analysis algorithms. Therefore, deep learning methods can be applied here to predict collisions by analyzing these factors in an LPWAN system. In this paper, we propose long short-term memory extended Kalman filter (LSTMEKF) model for collision prediction in the LPWAN in terms of the temporal correlation which can improve the LSTM performance. The efficacies of our models are demonstrated on the data set simulated by LoRaSim.