Distributed Precoding for eMBB and URLLC Traffic in Cell-Free O-RAN: A Multi-Agent Reinforcement Learning Framework

The integration of cell-free multiple-input multiple-output (MIMO) technology within the open radio access network (O-RAN) architecture addresses the growing need for decentralized, scalable, and high-capacity networks that can support different applications and use cases. In this paper, we propose a distributed precoding framework to support enhanced mobile broadband (eMBB) and ultra-reliable low-latency communications (URLLC) traffic in cell-free O-RANs, where each user is served by multiple open radio units (O-RUs). We consider short packet transmission in order to satisfy the latency requirements of URLLC traffic. We formulate a precoding optimization problem to maximize the aggregate throughput of eMBB users subject to the latency constraint of URLLC users. We propose a multi-agent deep reinforcement learning (DRL) algorithm to solve the formulated problem in a distributed manner. In particular, an actor-critic DRL agent is assigned to each O-RU. The actor determines the precoding matrices. The critic evaluates the actor's policy. The critics have global knowledge of all agents' policies, which stabilizes training and enables collaboration among agents. Simulation results show that the proposed algorithm provides an aggregate eMBB throughput improvement by up to 55.4 % when compared with three state-of-the-art baseline schemes.