Deep Reinforcement Learning Based Subcarrier Allocation With End‐To‐End Implicit CSI Acquisition in Multiuser OFDM Systems

Abstract

In multiuser orthogonal frequency division multiplexing (OFDM) systems, subcarrier allocation is a key means for improving the system capacity. During subcarrier allocation process, the channel state information (CSI) plays a key role. Recently, implicit CSI is considered as a possible alternative to replace explicit CSI. But under the condition of implicit CSI, the existing subcarrier allocation algorithms using explicit CSI need to be redesigned. In this article, subcarrier allocation with implicit CSI acquisition in frequency‐division duplex (FDD) OFDM systems is investigated. To adapt to the implicit CSI, a corresponding deep reinforcement learning (DRL) based proportional fairness (PF) subcarrier allocation algorithm is proposed to maximize the long term PF utility. Simulation results show that compared with the subcarrier allocation algorithms with explicit CSI acquisition such as LS/MMSE channel estimation and individual neural network (NN) based channel estimation and feedback, the proposed scheme achieves higher system sum rate and user fairness and gets close to the traditional PF algorithm with perfect CSI, which demonstrates the effectiveness of the proposed subcarrier allocation scheme with implicit CSI acquisition.