Reciprocally opportunistic spectrum access

The problem of opportunistic spectrum access in cognitive radio networks (CRNs) is considered in this paper. Especially, we discuss designing distributed learning protocols for non-cooperative secondary users (SUs) to efficiently utilize the spectrum opportunities. It is well known that the SUs’ selfish behaviors will result in a network collapse. Each SU is thus enabled to form conjectures over how the competing SUs would respond to its decision makings, and believes that they alter future spectrum access strategies in proportion to its own current strategy variation. The conjectural belief adapts in accordance with local observations. The interaction among SUs is modeled as a stochastic learning procedure, in which each autonomous SU behaves as an intelligent agent. In this way, all SUs can independently “learn” the behaviors of their competitors, optimize the opportunistic spectrum access strategies, and finally achieve the goal of reciprocity in CRNs. Based on the belief model, two learning mechanisms, i.e., the best response learning algorithm and the gradient ascent learning algorithm, are proposed to stabilize the opportunistic CRNs. We also theoretically prove that the SUs’ stochastic behaviors and beliefs converge to the conjectural equilibrium under some conditions rising from the learning procedure. Numerical results are provided to evaluate the performance of the proposed mechanisms, and show that the achieved system performance gain outperforms the multi-agent Q-learning scheme and the Nash equilibrium scheme when there are numerous SUs in the CRN.