中文摘要：

目前针对认知无线电网络中TCP协议的研究大多假设次用户的感知是完美的，且未综合考虑TCP协议参数和感知时间等因素对TCP性能的影响。针对上述问题，在TCP Westwood协议的基础上，提出一种基于传输预判的改进TCP协议，建立基于认知无线电网络的TCP吞吐量跨层模型。采用部分可观测马尔可夫决策过程对有感知误差的次用户频谱感知和接入过程进行建模，将其转换为信念状态马尔可夫决策过程，使用Sarsa（λ）算法对其进行求解，以在最大化TCP吞吐量的同时得到最优感知时间。仿真结果表明，与TCP Reno和TCP Newreno协议相比，使用该方案所得的TCP拥塞窗口值分别提高约42％和27％，平均吞吐量分别提高约5．7％和5．5％，当感知时间为0．2S时，所得的TCP平均吞吐量为最大值。

英文摘要：

Most of the existing studies about TCP protocol in Cognitive Radio（CR） network assume that the Secondary User（SU）＇s perception is perfect. These studies also do not consider the TCP protocol parameters and sensing time＇s influence on the TCP performance. To solve the above problems, this paper proposes an improved TCP protocol which has the transmission pre-judgment ability based on the TCP Westwood（TCPW） protocol. It builds the cross-layer TCP throughput model on the basis of CR networks. Due to the perception errors of secondary user, its spectrum sensing and access problem is modeled as a Partial Observable Markov Decision Process（POMDP） which is then converted to belief state Markov Decision Process（MDP）, with the Sarsa（2） algorithm to achieve the solution of Belief State Markov Decision Process（BMDP） model. It can achieve the optimal sensing time while TCP throughput has the maximum value. Simulation results show that the TCP congestion window value obtains with this scheme is about 42% more than using the TCP Reno protocol, and is about 27% more than using the TCP Newreno protocol. The average throughput obtains with this scheme is about 5.7% higher than using TCP Reno, and is about 5.5% higher than using TCP Newreno. When the sensing time is 0.2 s, the resulting TCP average throughput is the maximum.