A Random Forest-Based Prediction Method of Atmospheric Duct Interference in TD-LTE Networks

Abstract

Atmospheric ducts are horizontal layers that happens under certain weather conditions in the lower atmosphere. Radio signals guided in atmospheric duct tend to experience less attenuation and spread farther, i.e.hundreds of kilometers. In large-scale deployed TD-LTE network, atmospheric ducts make faraway downlink wireless signals propagate beyond the designed protection distance and interfere local uplink signals, thus cause severe outage probability. In this paper, we use the real network-side big data from the current operated TD-LTE network owned by China Mobile to predict atmospheric duct interference via machine learning. The results shows that using machine learning can get a good prediction performance for ADI. In addition, we compare the training time of different machine learning algorithm and find RF (Random Forest) has a short training time meanwhile maintains similar prediction accuracy and is thus more practical to operators.