A systematic framework for dynamically optimizing delay-sensitive wireless transmission

Delay sensitive applications need to overcome the service problems in dynamic environments with respect to both the multimedia source data (e.g., variable bit-rate) and the wireless channels (e.g., fading channel). This paper considers the problem of point to point transmission of scalable video coding over a fading channel. We formulate the rate adaptation challenge of WLAN multimedia networks as a Markov Decision Process and resolve this problem online based on reinforcement learning. The buffer state, channel state, and video state were considered as a joint state of system to maximize the average Quality of Service under delay constraints. To improve the convergence speed of learning, system's underlying dynamics were partitioned into a priori known and a priori unknown components. The proposed learning algorithm exploits known information about the system, so that less information needs to be learned compared with that in conventional reinforcement learning algorithms.