LAUSR

Demand-supply equilibrium is a preferable state for a cell, since resources are fully utilized and users' quality-of-experience is guaranteed. In principle, a price will be determined for utilizing a base station (BS) based on its load, and users will react to the prices of available BSs. In this paper, a deep reinforcement learning (DRL) based dynamic pricing method is proposed for the fully-decoupled RAN (FD-RAN). The proposed method exploits the separate control channel provided by the control BS, and can reach overall equilibrium for data BSs, with multi-link cooperative transmission inherently supported. We elaborately design the state, action and reward of DRL, and utilize several techniques. Simulations are conducted to demonstrate the stability, performance, and generalization of our proposed method.