AI-Enabled Centralized Spectrum Sharing

This chapter considers the AI-enabled centralized spectrum sharing in the wireless edge network, where multiple access points (APs) are deployed to serve users by reusing the same spectrum band. To alleviate the severe interference caused by spectrum reusing, advanced power control techniques are introduced to manage the interference and improve the sum-rate of the whole network. Conventional power control techniques require instantaneous global channel state information (CSI) to optimize the power control strategy. Nevertheless, instantaneous global CSI is impractical to acquire due to the fast-changing nature of the time-variable channel. To address this problem, this chapter leverages DRL to design a centralized learning algorithm catering to the power control. Specifically, by establishing a local DNN at each AP, we propose a multiple-actor-shared-critic (MASC) method to train all local DNNs in a centralized manner in the cloud. Then, each AP can determine the transmit power with the well-trained local DNN together with the local observations. Simulation results demonstrate that the performance of the proposed algorithm exceeds the benchmarks in terms of sum-rate and time complexity.