A Robustness Analysis of Learning-Based Coexistence Mechanisms for LTE-U Operation in Non-Stationary Conditions

The use of Long Term Evolution (LTE) in the unlicensed 5 GHz band, referred to as LTE-U, is a promising enhancement to increase the capacity of LTE networks and meet the requirements of future systems. This paper considers a Q-learning based Channel Selection strategy to decide the most appropriate channel to use for downlink traffic offloading in the unlicensed band as a mechanism to greatly facilitate the coexistence among several LTE-U and/or Wi-Fi systems in the same band. The focus is placed on analyzing the robustness of the proposed approach in front of non-stationary conditions in the wireless environment. Simulation results allow assessing quantitatively the capability of the proposed strategy to relearn proper solutions when changes in the environment occur. Furthermore, the analysis evaluates quantitatively how fast the learning process has to be compared to the variations in the environment in order to retain an LTE-U throughput performance very close to the optimum one.