On the Enabling of Efficient Coexistence of LTE With WiFi: A Machine Learning-Based Approach

Abstract

The recently proposed extension of the LTE operation to the unlicensed spectrum, known as LTE-Unlicensed (LTE-U), is not only expected to alleviate the congestion in the licensed band but is expected to result in an increase in the network capacity, as well. Unfortunately, such extension is challenged by a coexistence problem with wireless technologies operating in the unlicensed spectrum, especially Wi-Fi. Therefore, this article employs time series forecasting methods to enable efficient LTE coexistence with Wi-Fi. This is done by enabling the LTE-U Home eNodeB (HeNB) to avoid collisions with Wi-Fi by predicting the state of the unlicensed channels prior to using them. Specifically, this research proposes a recurrent neural network-based algorithm that utilizes Long Short Term Memory (LSTM) networks with time series decomposition to predict the state of the channels in the unlicensed spectrum. The authors investigate the performance of the proposed approach using extensive simulations. The results show that the proposed LSTM-based method outperforms the classical Listen Before Talk (LBT) and duty-cycling approaches in terms of improved coexistence of LTE-U with Wi-Fi.