Research On On-line Cross-layer Learning Technology For Cognitive Radio Network

Cognitive radio (CR) is a kind of new wireless communication technology,which can perceive the environment information, dynamically adjust the radiooperation parameters, in order to meet user’s requirements and improve thecommunication system performance. CR provides a feasible way to solvecontradictions between the lack of wireless spectrum resources and the dailygrowing business requirements, which has gradually become the focus in theacademic circles. The key technology for cognitive radio is learning, actiondecision and the environment parameter reconfiguration. The TCP throughput isone of the key factors to measure its performance in cognitive radio network.There are many researches On TCP throughput optimization. The typicalresearches are as follows:1. Researches based on markov process (MDP) model,this kind of method requires network with complete knowledge, but wirelessenvironment is changeable, partially observable, which have some limitations.2.Considering jointing physical layer, MAC layer to optimize the performance ofthe data link, but it ignores the TCP performance. This paper mainly analyzesand summarizes the CR technology research status, which includes networklearning, cognitive engine design, cross-layer parameters configuration etc.Aiming at different kinds of user’s QoS requirements, the correspondingcross-layer learning approaches is proposed to adapt the various wirelessenvironment, to improve the system throughput. The main thesis work issummarized as follows.Firstly, jointing the channel selection and modulation and coding types inphysical layer, frame length choice in data link layer, the cross-layer learningengine is raised to meet user’s Qos and optimize the system throughput. In orderto adapt to various wireless environment, this paper abstracts channel model intoa continuous markov decision process, and introduces the Q-learning algorithmto implement the learning of channel characteristics. To effectively utilizestorage space, the SVM regression model structure based mass Q functionapproximation is employed. A cross-layer parameters configuration is proposedto adapt the various environments, and this learning module can improve the end-to-end system throughput on the premise of meet the user’s QoS. UnderMatlab wireless communication platform, the simulation results show that thesystem in the absence of a part of the wireless environment under the premise ofprior knowledge, the system can find the optimal strategy though learning. TheSVM regression approximation Q value function can be implemented toconfigure the systems across the layer parameters, which can satisfy QoS optimalstate to effectively realize cognitive radio learning reconstruction.Secondly, this paper proposes a new Q-BMDP algorithm. Each user in CRnetwork decides the physical (PHY) modulation mode, transmission power andMAC layer channel access to find TCP optimal throughput. As users perceive theexistence of the error to the environment, in this article, we fomulate thisproblem as a part observable markov decision process (POMDP) framwork, andconvert it into the belief state MDP, to find the optimal strategy through Q valueiteration. Simulation results show that the system can study the optimal strategyin the dynamic wireless environment to improve the TCP throughput.