摘要
针对传统TCP拥塞窗口更新、控制机制所导致的Ad Hoc网络吞吐量下降的问题,利用机器学习算法来改善TCP在Ad Hoc网络中的性能。该方法利用确认帧的时间间隔,通过连续动作集(CALA)算法快速学习并估计当前网络链路中的拥塞状况,从而能够迅速调整TCP拥塞窗口。仿真实验表明:当Ad Hoc网络环境较好时,学习型TCP的吞吐量略优于TCP-Few、TCP-Reno协议,但在环境较差的情况下,学习型TCP的吞吐量远远优于TCP-Few和TCP-Reno协议。
In view of the traditional TCP congestion window updating and controlling induced by the decline of the throughput, the performance of TCP is improved by using machine learning algorithm in AdHoc. This method is used to the interval arrive time of the acknowledge （ ACK）, fast learns and estimates congestion in the network through continuous action-set learning automata（ CALA）, so that they can quickly adjust the TCP con- gestion window. The simulation results show that the throughput of learning TCP is slightly better than TCP-Few and TCP-Reno. But the throughput of learning TCP is much better than TCP-Few and TCP-Reno in bad link conditions.
出处
《电视技术》
北大核心
2014年第19期65-68,共4页
Video Engineering
基金
国家自然科学基金项目(61379005)
国防基础科研计划资助项目(B3120110005
B3120133002)
西南科技大学博士基金项目(12zx7127)
