MDP Based Cross-layer Scheduling For Delay Sensitive Services

The delay sensitive scheduling problem based on MDP is studied in this paper.In a multi-user scheduling system with single service based on TDMA, userscompete for transmission slots for accessing channel. Scheduler is responsible for thedecision that which user gets the transmission token and with what rates according tothe interactive transmission rates of all the users.. In the process of computing expectedtransmission rates, we analyze the properties of the state value function again and the socalled batch update online learning algorithm is used to solve the scheduling problem.Simulation shows that the batch update algorithm assures different delay constraintswell. Meanwhile, via function approximation, hundreds of update is replaced by tens ofupdate, which saves storage space and accelerates the convergence.Then, multi-traffic is adopted. Priorities are determined according to QoS of thefour traffic types defined in3GPP/3GPP2standard. Based on delay deadlines, bufferdynamics and time-varying channel state information, we provide amulti-user/multi-traffic access and transmission mechanism. Meanwhile, an onlinelearning algorithm is applicable to this system and the simulation shows that the accesscoefficient varies with different delay constraints. Meanwhile, the performance of highpriority service is better that that of low priority service.