基于博弈学习的动态频谱分配算法研究  

An Investigation of Dynamic Spectrum Allocation Algorithm Based on Game Learning

作　　者：徐浩漫[1] 唐伦[1] 陈前斌[1] 

机构地区：[1]重庆邮电大学通信与信息工程学院,重庆400065

出　　处：《计算机仿真》2010年第5期100-104,共5页Computer Simulation

基　　金：国家科技支撑计划(2008BAH30B10);新世纪优秀人才支持计划(NCET-06-0778);重庆市自然科学基金项目(CSTC;2008BB2413);重庆市教育委员会科学技术研究项目(KJ080522);重庆市教育委员会科学技术研究项目(KJ090501)

摘　　要：为了解决认知无线网络中的动态频谱分配问题,提高频谱利用率和避免干扰,提出了一种基于学习的库诺(Cournot)博弈模型,将授权用户对于空闲频谱的分配行为模拟为动态的博弈过程,并赋予授权用户学习的能力,通过对博弈过程中博弈者行为的学习和总结形成新的博弈策略,而且还比较了将最优反应学习算法和模拟退火算法应用到自适应博弈学习中系统的性能和用户的收益。仿真结果表明,两种算法均能够使授权用户通过学习达到策略的均衡,而有限理性下基于模拟退火的自适应博弈学习算法的鲁棒性更强,收敛性更好,且能够使授权用户获得更高的收益。With the special purpose of dynamic spectrum allocation on wireless cognitive network, a general coumot games framework is propose based on learning. The model of the behavior of the primary users who allocate their free spectrum to the second users as a collection of dynamic games process is set up. At the same time, we suppose that the primary users have the learning ability , that is, they can make strategic decisions by learning from other players in the games. We apply and compare the best -response learning algorithm and the adaptive learning algo- rithm based on simulated annealing. The simulation results show that both methods enable the primary users to stabilize their strategy , but the latter one is more steady and can give primary users a better profit.

关 键 词：动态频谱分配 库诺博弈 模拟退火 最优反应动态学习 

分 类 号：TN911.5[电子电信—通信与信息系统] TN92[电子电信—信息与通信工程]