Research On Channel Selection Mechanism Based On Multi-armed Bandit In Cognitive Network

As the wireless communication technology is developing rapidly, people's requirement for the wireless spectrum resource is increasing, which leads to the spectrum resources become scanty. Meanwhile, the wireless network environment is also becoming more complex, so it is difficult for cognitive users to obtain the information from the complex network environment and it will affect the spectrum utilization. Therefore, it is one of the current hot topics to study how to obtain the complete parameter information from the unknown network and take full advantage of the spectrum. In the thesis, the problem of the channel selection and the spectrum usage based on multi-armed bandit without network environment information for opportunistic spectrum access will be researched, and studies how to predict these channels quickly and accurately, then makes a reasonable and effective strategy for channel selection. This thesis mainly studies the following two aspects:Firstly, in the case of single user multi-channel, in order to solve the problem of how to search the optimal channel to maximize the system throughput under the unknown channel available statistics, the thesis puts forward an index strategy based on upper confidence bound with variance factor to explore the unknown channel environment. A variance factor is introduced into the confidence interval of the upper confidence bound index in the strategy and which can reflect the fluctuation of a series of the current reward, and real-time adjusts the exploration interval for reducing cost, then applies the historical information to estimate and select the optimal channel with maximum idle probability in next slot. By iterating the history information, this strategy can gradually approach the optimal channel and eventually will guide the cognitive users towards the direction of the maximum reward.Secondly, under the unknown network environment, the problem of channel selection for multi-user is studied. The channel states of primary users are modeled as independent and identically distributed multi-armed bandit, and a fair access method based on channel grouping is proposed. This scheme firstly adopts the distributed index strategy to explore and learn the unknown channel and acquire the channel-index, then according to the channel-index, these channels are uniformly divided into several groups, finally a cycle access strategy based on channel grouping is proposed, which can effectively solve the collision between users and make each user obtain a channel-group for data communication. Finally the simulation results show that the proposed scheme can not only effectively solve the collision between cognitive users, and increase the usage rate of these channels with small idle probability and improve the spectrum utilization, thus the learning regret value obtained grows slowly with logarithmic curve, but also the proposed scheme can reflect the fairness among cognitive users.