Foresighted resource scheduling in software-defined radio access networks

In a software-defined radio access network (Soft-RAN), the centralized network controller (CNC) and the wireless service providers (WSPs) are decoupled. Specifically, the CNC schedules resources over time for the mobile terminals (MTs) based on the value functions sent by the WSPs. While the strategic WSPs compete against each other on behalf of their MTs to optimize the long-term expected payoffs. The interactions among WSPs under a temporally changing SoftRAN form a non-cooperative stochastic game, which is regulated by the CNC using a Vickrey-Clarke-Groves pricing mechanism. However, due to the selfishness of WSPs, global network information is unattainable during the interacting process. We hence propose a stochastic resource scheduling scheme, in which each WSP needs neither other WSPs´ private information nor a priori knowledge of network dynamics, yet it is able to independently learn the approximated true value functions in a gradual way. Numerical simulations are carried out to demonstrate the performance gains that can be achieved from the proposed scheme.