Network selection in cognitive heterogeneous networks using stochastic learning

Coexistence of multiple radio access technologies (RATs) is a promising paradigm to improve spectral efficiency. This letter presents a game-theoretic network selection scheme in a cognitive heterogeneous networking environment with time-varying channel availability. We formulate the network selection problem as a noncooperative game with secondary users (SUs) as the players, and show that the game is an ordinal potential game (OPG). A decentralized, stochastic learning-based algorithm is proposed where each SU's strategy progressively evolves toward the Nash equilibrium (NE) based on its own action-reward history, without the need to know actions in other SUs. The convergence properties of the proposed algorithm toward an NE point are theoretically and numerically verified. The proposed algorithm demonstrates good throughput and fairness performances in various network scenarios.