Reinforcement learning cooperative congestion control for multimedia networks

A cooperative congestion control based on the learning approach to solve congestion control problems on multimedia networks is presented. The proposed controller, which is capable of rate-based predictive control, consists of two sub-systems: a long-term policy critic and a short-term rate-adaptor. Each controller in a chained network jointly learns the control policy by real-time interactions without prior knowledge of a network model. Furthermore, a cooperative fuzzy reward evaluator provides cooperative reinforcement signals based on game theory to train controllers to adapt to dynamic network environment. The well-trained controllers can take correct actions adaptively to regulate source flow to simultaneously meet the requirements of high link utilization, low packet loss rate (PLR) and end-to-end delay. Simulation results show that the proposed approach is very effective in controlling congestion of the multimedia traffic in Internet networks.