Time series forecasting for rural fixed-wireless communication network monitoring

Thumbnail Image



Journal Title

Journal ISSN

Volume Title


University of New Brunswick


LTE and 5G cellular networks are evolving at a rapid pace to accommodate more users and higher traffic. Existing studies have largely focused on urban mobile networks, leaving their rural fixed-wireless counterparts largely ignored. This work investigates the performance of a rural Canadian fixed wireless network on several time scales. Short- and long-term performance properties are considered. It is well known that rural propagation environments behave differently from urban ones. Long-term temporal changes in the propagation environment, such as foliage and snow, were shown to have a small impact on the performance of the network. From a forecasting point of view, it was shown that including environmental features and increasing the time horizon of the forecasts will increase the accuracy of the forecast. In contrast, it was shown that including environmental features did not provide any benefit to short-term forecast accuracy; however, longer input sequence lengths were demonstrated to be beneficial for short-term forecasts. Finally, an unsupervised anomaly detection algorithm, RAINFOREST, is presented which leverages the temporal context obtained from the forecasts alongside density-based clustering analysis to outperform all the baselines tested.