A Recommendation System in Cognitive Radio Networks With Random Data Traffic

A recommendation system is proposed to enhance the efficiency of spectrum access in cognitive radio networks, in which secondary users broadcast the indices of channels that they have successfully accessed. The probabilities of different actions, i.e., taking a recommendation or probing an unrecommended channel, could be either fixed or adjustable. For the constant probability case with and without retransmission, the system is modeled as a Markov random process, and the corresponding state transition probabilities are obtained. For the adjustable probability case, the anytime multiarmed bandit technique is used to adopt the strategies to the uncertain environments, and a performance lower bound is obtained. Numerical simulation results demonstrate that the proposed recommendation system can effectively orient the channel selections and significantly improve the performance of cognitive radio networks.