Research On Access Strategy Of Energy Harvesting Cognitive Radio Network Under Strict Transmission Time Limi

In recent years,spectrum resources have become overly crowded due to the rapid development of wireless communication services,and thus cognitive radio technology has received much research attention owning to its efficient spectrum utilization.In general,secondary users(SUs),which are allowed to opportunistically access the spectrum that is unoccupied by primary users(PUs),needs to jointly take into account the activity patterns of PUs,the strict delivery deadline constraint required by real-time data services and the energy harvesting constraint due to the lack of fixed energy supply.The mutual couplings of these three factors make the channel access design of SUs more challengeable.However,previous studies only considered fixed access parameters or did not have a full consideration of the time-varying and partially-observable characteristic of environmental information.To overcome this weakness,this thesis aims to design and optimize the channel access strategy of SUs for the following three cases.(1)For the single-channel case with previously known environmental parameters(e.g.,packet arrival rates,channel transition probability matrix and packets successful transmission probabilities),by considering time-varying and completely observable channel status,we use an infinite-horizon Markov decision process(MDP)to provide a completed modeling for the channel access of the SU relying on the urgencies of all packets,and also use an infinite-horizon MDP to provide a simplified modeling only relying on the urgency of head-of-queue(Ho L)packet.We further prove the unichain property of these two MDPs,and apply the value iterative algorithm to obtain deterministic stationary ε-optimal policies for these two MDPs.Simulation results demonstrate the effectiveness of simplified model and the performance advantage of the proposed strategies over other strategies.(2)For the multichannel case with previously known environmental parameters,by considering time-varying and partially observable channel statuses,we use an infinite-horizon partially observable Markov decision process(POMDP)to provide a simplified modeling only relying on the urgency of Ho L packet,obtain a near-optimal strategy using the exact value iterative algorithm for the finite-horizon problems,and further obtain a suboptimal strategy using the Q-function Markov decision process(QMDP).Simulation results demonstrate the performance advantage of the proposed strategy over other strategies.(3)For a more practical case without any prior knowledge of environmental parameters,relying on environmental feedback such as sensed channel status and packet transmission outcome,we design an online learning model based on the proposed simplified MDP modeling for the single-channel scenario,design another online learning model based on the proposed simplified POMDP modeling for the multichannel scenario.We further use policy gradient estimation algorithms to obtain online learning strategies for these two scenarios,respectively.Simulation results demonstrate the effectiveness of the proposed learning strategy and the performance advantage over other strategies.