Distributed Deep Reinforcement Learning for Radio Resource Management in O-RAN
Ahmad Ahmadi Siahpoush, Vahid Shah‐Mansouri · 2024
The open radio access network (O-RAN) has been developed with the purpose of enabling intelligence and openness in next generation cellular networks. It relies on virtualized, disaggregated, and software-driven components which provides an open environment for network vendors and operators. O-RAN offers standardized interfaces and supports the hosting of network applications from third-party vendors through x-applications (xApps), thereby enhancing network management flexibility through the utilization of artificial intelligence (AI) and machine learning (ML) techniques. The resource management in O-RAN is performed by xApps implemented in the near-real-time RAN intelligent controller (Near-RT RIC). In this Paper, xApps are modeled as deep reinforcement learning (DRL) agents that perform optimal radio resource management through interaction with each other and with the O-RAN environment. In particular, each xApp is considered to be a deep Q-network (DQN) agent that is responsible for the joint optimization of power and radio re-source allocation. We also propose a distributed radio resource management algorithm where multiple xApps collaborate in a distributed manner, aiming to minimize interference between network users and enabling them to reach their maximum data transmission capacity. Our experimental results show the superior performance of the proposed algorithm for the distributed management of radio resources compared to a decentralized approach, while also achieving comparable performance with a centralized approach.