Machine Learning based Atmospheric Duct Interference Evaluation in TD-LTE Networks

Atmospheric ducts are natural channels in Earth's lower atmosphere that allow electromagnetic signals to travel long distances. However, in short-distance communication systems like Time Division Duplexing Long Term Evolution (TD-LTE) networks, these ducts can cause atmospheric duct interference (ADI). The downlink signals of one base station tend to interfere with the uplink signals of another base station as the electromagnetic waves of the base station travel beyond its planned radius due to atmospheric ducts. This can negatively affect the Quality of Service (QoS) and Service Level Agreements (SLA) in TD-LTE networks. Hence, this work proposes a machine learning-based ADI detection and mitigation methodology that adaptively changes channel parameters to avoid ADI. First, based on the real-world data set, we develop a model to detect ADI in advance. These data are analysed using three mostly cited algorithms in the literature on detecting ADI: Support Vector Machine (SVM), Random Forest, and Long Short-Term Memory (LSTM), and their performance are analysed to investigate the best feasible model considering the algorithm complexity, real-time capability, accuracy, precision, and recall of the predicted results. The support vector machine shows the highest test accuracy among the three models, with a test accuracy rate of 63.8% in detecting ADI. This performance surpasses the results reported in the literature. In addition, a technique is suggested for mitigating ADI based on the guard period, which refers to the time interval between uplink and downlink signals in the fourth-generation long-term evolution (4G-LTE) network. The proposed method allows for adjusting the guard period based on real-time atmospheric conditions and network-related data, reducing the impact of atmospheric ducting on the network's performance.