A Proximity-Based Q-Learning Reward Function for Femtocell Networks

Q-learning has become a prominent tool for resource allocation and management in femtocell networks, as it can decrease the time the network takes to determine an allocation of resources by sharing information. The choice of reward function in Q-learning can greatly affect the performance of the Q-learning algorithm in the resource allocation process. In this work, we present a novel reward function for Q-learning that takes into consideration the femtocell proximity to mobile users. Components of the reward function are emphasized and de-emphasized as a function of distance, based on the influence of the agent on each component. In comparison with other methods, the proposed method ensures a certain level of service for the primary user and increases the sum capacity of the network.