Research On Dynamic Spectrum Access In Cognitive Radio

Since spectrum resources are becoming more significant and rarer inwireless communication, researchers proposed dynamic spectrum access toimprove efficiency of spectrum usage. In dynamic spectrum access, usersare allowed to share spectrum in various methods, which is calledspectrum access policy. It’s very common that different network usesdifferent spectrum access policy. In wireless communication system,getting an appropriate spectrum access policy is very crucial, which affectsnot only the throughput of network but also the QoS.In this paper, we focus on a scenario in which users are not informedabout the channel quality in network. So the users must learn about thenetwork in order to maximize rewards. Our research consists of two parts.In the first part, we consider a wireless network with only one user.The novel point is that we take switch cost into consideration. We supposethe operation of switch will reduce the user’s rewards due to the time delayand extra energy consumption. We model this problem as the multi-armedbandit problem. Firstly, a modified Lai-Robbins policy is proposed. Wefind the regret under modified Lai-Robbins policy grows with largeleading constant due to the frequent channel switch. Then we proposed theRSAP (Reduce Switch with Advanced Play) policy, under which the regretis proved to grow with much smaller leading constant.Since heterogeneous network has become the trend of future wirelesscommunication system, in the second part, we propose a distributedlearning algorithm for heterogeneous wireless network. In heterogeneousnetwork, the situation is very complicated because of the following reasons.First, it’s a distributed system, so there is no control center. Second, the payoff functions are various depending on different types of base stations.Third, users do not know the payoff functions exactly, since channelquality and actions of others are not known. We model the dynamicspectrum access problem in heterogeneous network as the ASPS(Asymmetric Singleton congestion game with Player-Specific payofffunctions). It’s proved that the ASPS has Nash equilibrium. Then wepropose the better-reply distributed learning algorithm. We can prove thatunder the better-reply distributed learning algorithm, the ASPS problemwill converge to Nash equilibrium.