Spectrum decision in cognitive radio networks using multi-armed bandit

In cognitive radio networks, secondary users utilize idle parts of primary users' spectrum in order to achieve more spectrum efficiency. However, due to the fact that sensing more than one channel at every sensing opportunity is difficult, the spectrum selection by the secondary users is a serious challenge in real senarios. In this paper, the problem of spectrum sensing and selection in a cognitive radio system is modeled as a multi-armed multi-player bandit problem. Unlike to the other works in this area, we have considered a bursty type traffic for primary users and by introducing a forgetting factor to the involved past successful transmissions in the bandit model, it is shown that the average successful transmission rate is increased. Simulation results show that successful transmission rate in our algorithm is 10% better than that of random spectrum sensing.