An Improved Call Admission Control Scheme Based on Reinforcement Learning for Multimedia Wireless Networks

This paper presents an improved call admission control scheme to optimize the network operators´ revenue while guarantying the quality of service (QoS) to the mobile terminals. The problem of call admission control (CAC) is modeled as a Semi-Markov decision process (SMDP), and the SMDP is solved by a reinforcement learning (RL) algorithm known as Q-learning. In the Q-learning algorithm, the reward functions for the acceptance and the rejection of new calls for each class of service not only depend on used bandwidth, new call arrival rate, average service time and price, but also the ratio of new call load and the handoff call load and the requested bandwidth of each class of traffic. The CAC scheme would be well performed through the reward functions. Simulations results show that the CAC scheme can obtain high revenue while greatly reducing handoff call dropping probability when the traffic loads are heavy.