GTDM-CSAT: an LTE-U self Coexistence Solution based on Game Theory and Reinforcement Learning

Abstract

There is substantial literature covering both problems and solutions related to the operation of Long Term Evolution (LTE) networks in unlicensed spectrum (LTE-U) while in coexistence with other technologies, such as Wi-Fi. However, a seldom explored scenario is the coexistence between multiple LTE-U networks. Within this scenario, a big issue is establishing optimal configurations that take into account fairness among different operators coexisting in the same unlicensed spectrum coverage area. Solutions to this problem could, for instance, react to changes in the environment by "tuning" different system configurations. We propose a game theoretical reinforcement learning algorithm, called GTDM-CSAT, aiming to maximize the LTE-U aggregated throughput while keeping channel access fairness among different access points. GTDM-CSAT uses the relative data rate offered by the system to adapt the LTE-U ON-OFF time. For this, we formulate the problem as a Markovian game, where the LTE-U operators coexist on a two-zero-sum game. The solution for the best ON-OFF time ratio is defined by applying a modified Minimax Q-learning algorithm for finding the game equilibrium. We perform simulations following 3GPP specifications using the ns-3 simulator for evaluating GTDM-CSAT under different traffic load scenarios. Results indicate gains in the system aggregated throughput by improving the individual data rate of each operator.

Solution for Interference in Hotspot Scenarios Applying Q-Learning on FFR-Based ICIC Techniques

“…Our previous works include [ 43 , 53 ]. They present solutions that use ABS and QL to coordinate the shared transmission between LTE and Wi-Fi in the 5 GHz bandwidth.…”

A Machine Learning Approach for Handover in LTE Networks with Signal Obstructions

“…Machine Learning has been vastly used to improve the networks' performance, especially in the last few years [38]- [41]. The authors of [38] propose a method based on Deep Neural Networks to mitigate the link failure caused by unsuccessful handovers and congested cells, among other reasons.…”

“…Aiming to detect network intrusion, the authors of [42] develop a technique based on Random Forest (RF) and Support Vector Machine (SVM), while the researchers of [39] use an approach based on Deep Learning. Solutions for coexistence in the unlicensed band are proposed in [41] for LTE/Wi-fi networks, and in [40] for LTE-U systems. The authors demonstrate meaningful gains and the flexibility to dynamically adjust the system parameters in response to the varying behavior of the inter-system interference.…”

A Multiarmed Bandit Approach for LTE-U/Wi-Fi Coexistence in a Multicell Scenario

“…In our previous works [ 6 , 7 , 8 , 9 , 10 ], we have demonstrated how the first two solutions work and how they interact with Wi-Fi, which is currently the most successful technology utilizing the unlicensed spectrum. We also proposed a centralized Q-Learning (QL) solution to increase system throughput in a coexistence multicell scenario with LTE-U and Wi-Fi [ 10 ].…”