A model based RL admission control algorithm for next generation networks

In this paper we study the call admission control problem to optimize the network operators' revenue guaranteeing the quality of service to the end users. We consider a network scenario where each class of service is characterized by a different constant bit rate and an associated revenue. We formulate the problem as a Semi-Markov Decision Process, and we use a model based Reinforcement Learning approach. Other traditional algorithms require an explicit knowledge of the state transition models while our solution learns it on-line. We will show how our policy provides better solution than a classic greedy algorithm.