Deep-Reinforcement-Learning-Based Scheduling with Contiguous Resource Allocation for Next-Generation Cellular Systems

In this work, we propose a novel scheduling algorithm with contiguous frequency-domain resource allocation (FDRA) based on deep reinforcement learning (DRL) that jointly selects users and allocates resource blocks (RBs). The scheduling problem is modeled as a Markov decision process, and a DRL agent determines which user and how many consecutive RBs for that user should be scheduled at each RB allocation step. The state, action, and reward sets are delicately designed to train the DRL network. More specifically, the originally quasicontinuous action space, which is inherent to contiguous FDRA, is refined into a finite and discrete action space to obtain a tradeoff between the inference latency and system performance. Simulation results show that the proposed DRL-based algorithm outperforms other representative baseline schemes while having lower online computational complexity.