Reinforcement learning for control of self-similar call traffic in broadband networks

Reinforcement learning is applied to admission control of self-similar call traffic in broadband networks. The reinforcement learning method solves a Markov Decision Problem without the need for a model of the dynamics of the controlled system. A state descriptor containing continuous-valued running averages of the call inter-arrival times is employed. Radial-basis function neural networks approximate the value function. In simulations, the proposed method yields higher throughput than methods that do not exploit the self-similarity of the call arrival process.