Distributed Q-learning based-Decentralized Resource Allocation for Future Wireless Networks

The next generation (5G) wireless network is expected to support the fourth industrial revolution, combining heightened data transfer speeds and processing power. Despite technological progress, the most important evolution of future networks is the network-densification technique. This technique reduces significantly the propagation distance and makes signal propagation transit from long- to short-range propagation. In this regard, several imposed challenges such as mobility and interference management, and network and resource management can drain the spatial resources and degrade network performance and end user Quality of Service (QoS), when the network density exceeds a critical density. We focus in this paper on the resources management and allocation in the context of exploiting the spatial reuse. However, Nodes are authorized to modify their transmission power (TP) and channel according to the resulting throughput. Toward this goal, a distributed Q-learning, from the Reinforcement learning category (RL), is proposed to allow nodes to select their TP and the channel only depending on their bit rate rewards. Simulation result proves, after having evaluated the effect of the learning parameters, the effectiveness of the proposed method in finding the best-performing configurations in terms of throughput maximization.