Abstract: Packet scheduling technology is one of the important technical issues in designing wireless multimedia networks. The key difficulties lie in the high variability of wireless channel capacity and the unknown model of packet arrival processes. This paper views the packet scheduling problem as a Semi-Markov Decision Process(SMDP), and approximately solves the SMDP problem by the methodology of Neuro-Dynamic Programming(NDP). The proposed algorithm, called Neuro-Dynamic Programming Scheduling(NDPS), can simultaneously achieve three performance objectives: QoS differentiation and guarantee, high bandwidth utilization, and both short-term and long-term fairness. Simulation experimentalresult demonstrates that NDPS outperforms two popular scheduling algorithms.

Key words: packet scheduling, multimedia networks, Markov decision processes, neuro-dynamic programming

摘要： 分组调度是实现未来无线多媒体网络的关键技术之一。解决该技术的主要困难在于无线链路的高差错率、业务类型的多样性和分组到达模型的未知性。该文引入马尔可夫决策过程对分组调度过程进行建模，运用神经动态规划方法求解相应的马尔可夫决策过程问题。提出一种无线多媒体网络分组调度算法(NDPS)可以同时实现3个性能目标：对不同业务类型提供差分服务，最大化无线带宽的利用率和保证服务公平性。仿真实验结果证明，NDPS算法比两种流行的调度算法具有更好的性能。

关键词: 分组调度, 多媒体网络, 马尔可夫决策过程, 神经动态规划

CLC Number: 

• TP33