Wireless Network Association Game With Data-Driven Statistical Modeling
Abstract: The explosion in demand for wireless data services in recent years has triggered pervasive deployment of wireless networks. How to associate to one of the wireless networks in the best interest of a user is an essential problem to mobile computing. In this paper, we analyze a data set of wireless LAN traces collected from campus networks, from which we observe that the user arrival distribution is approximately Poisson distributed; the session time and the waiting time to switch network can be approximated by exponential distributions. Based on the data analysis, we formulate a wireless access network association game as a multidimensional Markov decision process with negative network externality, where the best response strategy is an approximate Nash equilibrium. A modified value iteration algorithm is proposed to search the best response strategy profile. Applying the proposed algorithm to the data-driven stochastic model, the best response strategy is shown to achieve a better individual expected utility while satisfying the individual rationality, and attain a near-optimal social welfare performance compared to other strategies such as the centralized method and the greedy algorithm.