Abstract
We present a novel game, called evolutionary Bayesian coalition formation game, to model and analyze the problem of distributed resource sharing in a multi-cell device-to-device (D2D) enabled cellular network where the rationality of the players, i.e., device pairs, is bounded, e.g., due to limited information. Each player can make its decision on the channel to access with and without coordination. In the former case, the player works in D2D mode. In the latter case, the player forms a coalition with some other players and they connect to one base station in cellular mode. In this case, the player realizes its action after observing the actions of other players. Unlike classical coalition formation games where the player decides on its coalition to form by estimating its payoff, in the proposed game, the player forms a coalition and selects an action based on its current population state which is updated using a simple and scalable learning algorithm. We prove that the evolutionary coalition formation process converges to the unique equilibrium that induces a stable coalitional agreement. The proposed process is applied to a long-term evolution-advanced network where it shows a superior performance compared with other baseline resource sharing strategies.